Is OCaml fast?

Is OCaml fast?

[Thanassis Tsiodras]

I apologize beforehand if this is not the forum to ask.

I am on the fence about whether to learn OCaml or not, and while
reading an article called "Why OCaml"
(http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw
that OCaml was praised for the speed of the executables it generates -
and was referred to, speed-wise, as "second to none", except C and
C++.

However, when I actually went to the Language Shootout page suggested
in the article, I found out that OCaml is not 2nd, it is 13th, behind
languages like Haskell and C#...
(http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)

Is it just hype, then? Or am I missing something?

-- 
What I gave, I have; what I spent, I had; what I kept, I lost. (Old Epitaph)

Re: [Caml-list] Is OCaml fast?

[Gregory Bellier]

[-- Attachment #1: Type: text/plain, Size: 1525 bytes --]

If you want speed then you should learn assembly or choose C as a second
choice.
It depends on the kind of apps you'd like to write. Even though O'Caml is
fast, it's not the first criteria I have in mind which would be security :
no segfault, no need to handle horrible stuff like in C, ...


2010/11/22 Thanassis Tsiodras <ttsiodras@gmail.com>

> I apologize beforehand if this is not the forum to ask.
>
> I am on the fence about whether to learn OCaml or not, and while
> reading an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html<http://www.cs.ubc.ca/%7Emurphyk/Software/Ocaml/why_ocaml.html>),
> I saw
> that OCaml was praised for the speed of the executables it generates -
> and was referred to, speed-wise, as "second to none", except C and
> C++.
>
> However, when I actually went to the Language Shootout page suggested
> in the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
> (
> http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php
> )
>
> Is it just hype, then? Or am I missing something?
>
> --
> What I gave, I have; what I spent, I had; what I kept, I lost. (Old
> Epitaph)
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>

[-- Attachment #2: Type: text/html, Size: 2320 bytes --]

Re: [Caml-list] Is OCaml fast?

[Lukasz Stafiniak]

On Mon, Nov 22, 2010 at 2:35 PM, Gregory Bellier
<gregory.bellier@gmail.com> wrote:
> If you want speed then you should learn assembly or choose C as a second
> choice.
> It depends on the kind of apps you'd like to write. Even though O'Caml is
> fast, it's not the first criteria I have in mind which would be security :
> no segfault, no need to handle horrible stuff like in C, ...

No! You should use ATS, just as the shootout suggests. (Portable)
assembly is not a language for humans ;-)

Re: [Caml-list] Is OCaml fast?

[Thomas Fischbacher]

Gregory Bellier wrote:

> If you want speed then you should learn assembly or choose C as a second 
> choice.

Certainly not assembly. Modern microprocessors just see assembly as a
kind-of high level language which they interpret in a funny way, doing
all sorts of re-schedulings, register renamings, delayed branching,
speculative execution etc. Producing code that can take advantage of
this is best left to the compiler - unless one has to deal with
instruction set extensions for which there is no good compiler
support yet.

> It depends on the kind of apps you'd like to write. Even though O'Caml 
> is fast, it's not the first criteria I have in mind which would be 
> security : no segfault, no need to handle horrible stuff like in C, ...

I'd say the OCaml native compiler is reasonably fast for pretty much all
applications that need the speed of compiled code. But I'd say the same
for GHC and SBCL, say. There are some good reasons to take a closer look
at OCaml, but these are related to other qualities of the language.

In my view, its greatest benefit is that it makes working with closures
simple while being fast and by far not as intimidating to (prospective)
PhD students as Scheme/Lisp (but that only because most of them have
been spoilt by imperative languages beforehand).

-- 
best regards,
Thomas Fischbacher
t.fischbacher@soton.ac.uk

Re: Is OCaml fast?

[Sylvain Le Gall]

Hello,

On 22-11-2010, Thanassis Tsiodras <ttsiodras@gmail.com> wrote:
> I apologize beforehand if this is not the forum to ask.
>
> I am on the fence about whether to learn OCaml or not, and while
> reading an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw
> that OCaml was praised for the speed of the executables it generates -
> and was referred to, speed-wise, as "second to none", except C and
> C++.
>
> However, when I actually went to the Language Shootout page suggested
> in the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
> (http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)

Shootout benchmarks doesn't always allow to set some variables that
could greatly improve the speed of OCaml.

>
> Is it just hype, then? Or am I missing something?
>

You could write very fast application with OCaml -- even beating C code
in some case. But what OCaml is really helpful for, is that you can do
algorithmic optimizations that improve the speed. It means that OCaml
is terse enough to allow you to write complex algorithm without making
too much bugs. 

All in all: development time + execution time is fast.

But if you spend 10x development time on the same C code, you will
obviously get something faster in C. 

Regards,
Sylvain Le Gall

Re: [Caml-list] Is OCaml fast?

[Dario Teixeira]

Hi,

> I am on the fence about whether to learn OCaml or not, and while
> reading an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html),
> I saw that OCaml was praised for the speed of the executables it
> generates - and was referred to, speed-wise, as "second to none",
> except C and C++.

Yes, Ocaml is fast (more on that later), but you should consider speed
as just the icing in the cake.  Besides all the cool language features,
what makes me recommend Ocaml to people who -- like you -- may be on
the fence, is that among the not so mainstream languages, Ocaml is
perhaps the one people are more likely to stick with for actual work
after they get past the learning stage.  Remember that though Ocaml
favours the functional paradigm, it is flexible enough to support those
(few) situations where it actually makes sense to be imperative and/or
use an object oriented approach, and where a purely functional solution
is just intellectual masturbation.

> However, when I actually went to the Language Shootout page suggested
> in the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
> (http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)
> 
> Is it just hype, then? Or am I missing something?

There's lies, damn lies, and shootout statistics.  Getting good performance
in Haskell is often something of a black art that baffles even gurus. In
contrast, Ocaml has very good "performance predictability".  Moreover, it
is possible to get good performance from fairly idiomatic Ocaml programmes,
whereas in other languages you may need to write ugly code which looks like
nothing one would write for elegance and sanity.

Best regards,
Dario Teixeira

Re: [Caml-list] Is OCaml fast?

[Gerd Stolpmann]

Am Montag, den 22.11.2010, 15:21 +0200 schrieb Thanassis Tsiodras:
> I apologize beforehand if this is not the forum to ask.
> 
> I am on the fence about whether to learn OCaml or not, and while
> reading an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw
> that OCaml was praised for the speed of the executables it generates -
> and was referred to, speed-wise, as "second to none", except C and
> C++.
> 
> However, when I actually went to the Language Shootout page suggested
> in the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
> (http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)
> 
> Is it just hype, then? Or am I missing something?

I think the shootout is not a good data source. There are definitely
some very poor Ocaml results there, so I'd guess the shootout got
recently more attention by enthusiasts of other languages, and the
current Ocaml programs there are not very good. (I remember Ocaml was #1
at the shootout a few years ago, faster than C.) So maybe a good
opportunity to post better Ocaml solutions there?

Gerd
-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

[was: Re: Is OCaml fast?] OCaml Shootout task force

[Sylvain Le Gall]

On 22-11-2010, Gerd Stolpmann <info@gerd-stolpmann.de> wrote:
> Am Montag, den 22.11.2010, 15:21 +0200 schrieb Thanassis Tsiodras:
>> I apologize beforehand if this is not the forum to ask.
>> 
>> I am on the fence about whether to learn OCaml or not, and while
>> reading an article called "Why OCaml"
>> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw
>> that OCaml was praised for the speed of the executables it generates -
>> and was referred to, speed-wise, as "second to none", except C and
>> C++.
>> 
>> However, when I actually went to the Language Shootout page suggested
>> in the article, I found out that OCaml is not 2nd, it is 13th, behind
>> languages like Haskell and C#...
>> (http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)
>> 
>> Is it just hype, then? Or am I missing something?
>
> I think the shootout is not a good data source. There are definitely
> some very poor Ocaml results there, so I'd guess the shootout got
> recently more attention by enthusiasts of other languages, and the
> current Ocaml programs there are not very good. (I remember Ocaml was #1
> at the shootout a few years ago, faster than C.) So maybe a good
> opportunity to post better Ocaml solutions there?
>

Maybe it is time to join forces and update the shootout for OCaml?

If some people are interested here, I can setup a repository on the forge
to update all this tests. 

If I get 2 people interested, I will setup a VCS repository + project on
the forge, this afternoon. Drop me an email + preferred VCS (among svn,
git, darcs) and your account login on http://forge.ocamlcore.org. I will
take care, when ready, to made this code available in the shootout once
finished.

Regards,
Sylvain Le Gall

Re: [Caml-list] Is OCaml fast?

[bluestorm]

[-- Attachment #1: Type: text/plain, Size: 1700 bytes --]

On Mon, Nov 22, 2010 at 3:04 PM, Gerd Stolpmann <info@gerd-stolpmann.de>wrote:

> I think the shootout is not a good data source. There are definitely
> some very poor Ocaml results there, so I'd guess the shootout got
> recently more attention by enthusiasts of other languages, and the
> current Ocaml programs there are not very good. (I remember Ocaml was #1
> at the shootout a few years ago, faster than C.) So maybe a good
> opportunity to post better Ocaml solutions there?
>

As Sylvain noticed, some (in not most) of the OCaml poor performances in the
shootout are actually not due to bad OCaml programs, but to arbitrary
restrictions in the shootout rules. For example, one of the bad-performing
benchmark for OCaml is the binary-tree benchmark, where it is nearly four
times slower than C, but on closer inspection you discover that this is due
to the arbitrary choice to forbid any change of the GC parameters. With
appropriate GC parameters, the very same OCaml program is exactly as fast as
C.

http://shootout.alioth.debian.org/u32/performance.php?test=binarytrees

« Note: these programs are being measured with *the default initial heap
size* - the measurements may be very different with a larger initial heap
size or GC tuning. »
C version : 12.11 secs
OCaml version : 47.22 secs
OCaml version with GC parameters tuned ("interesting alternative" section) :
12.67 secs


Therefore, there is nothing that can be changed to the OCaml submission for
this benchmark to improve performances, except changing the default GC
parameters; while this might be a good idea in general, changing it only for
the sake of shootout-obsessed people is ridiculous.

[-- Attachment #2: Type: text/html, Size: 2691 bytes --]

Re: [Caml-list] Is OCaml fast?

[Török Edwin]

On Mon, 22 Nov 2010 15:36:30 +0100
bluestorm <bluestorm.dylc@gmail.com> wrote:

> On Mon, Nov 22, 2010 at 3:04 PM, Gerd Stolpmann
> <info@gerd-stolpmann.de>wrote:
> 
> > I think the shootout is not a good data source. There are definitely
> > some very poor Ocaml results there, so I'd guess the shootout got
> > recently more attention by enthusiasts of other languages, and the
> > current Ocaml programs there are not very good. (I remember Ocaml
> > was #1 at the shootout a few years ago, faster than C.) So maybe a
> > good opportunity to post better Ocaml solutions there?
> >
> 
> As Sylvain noticed, some (in not most) of the OCaml poor performances
> in the shootout are actually not due to bad OCaml programs, but to
> arbitrary restrictions in the shootout rules. For example, one of the
> bad-performing benchmark for OCaml is the binary-tree benchmark,
> where it is nearly four times slower than C, but on closer inspection
> you discover that this is due to the arbitrary choice to forbid any
> change of the GC parameters. With appropriate GC parameters, the very
> same OCaml program is exactly as fast as C.
> 
> http://shootout.alioth.debian.org/u32/performance.php?test=binarytrees
> 
> « Note: these programs are being measured with *the default initial
> heap size* - the measurements may be very different with a larger
> initial heap size or GC tuning. »
> C version : 12.11 secs
> OCaml version : 47.22 secs
> OCaml version with GC parameters tuned ("interesting alternative"
> section) : 12.67 secs
> 
> 
> Therefore, there is nothing that can be changed to the OCaml
> submission for this benchmark to improve performances, except
> changing the default GC parameters; while this might be a good idea
> in general, changing it only for the sake of shootout-obsessed people
> is ridiculous.

Isn't it possible for the GC to realise its doing too many collections
and increase the minor heap size on its own?
Or isn't it possible to determine at compile time an approximation for
a good heap size, and use that?

Best regards,
--Edwin

Re: [Caml-list] Is OCaml fast?

[Gerd Stolpmann]

Am Montag, den 22.11.2010, 15:36 +0100 schrieb bluestorm:
> On Mon, Nov 22, 2010 at 3:04 PM, Gerd Stolpmann
> <info@gerd-stolpmann.de> wrote:
>         I think the shootout is not a good data source. There are
>         definitely
>         some very poor Ocaml results there, so I'd guess the shootout
>         got
>         recently more attention by enthusiasts of other languages, and
>         the
>         current Ocaml programs there are not very good. (I remember
>         Ocaml was #1
>         at the shootout a few years ago, faster than C.) So maybe a
>         good
>         opportunity to post better Ocaml solutions there?
> 
> 
> As Sylvain noticed, some (in not most) of the OCaml poor performances
> in the shootout are actually not due to bad OCaml programs, but to
> arbitrary restrictions in the shootout rules. For example, one of the
> bad-performing benchmark for OCaml is the binary-tree benchmark, where
> it is nearly four times slower than C, but on closer inspection you
> discover that this is due to the arbitrary choice to forbid any change
> of the GC parameters. With appropriate GC parameters, the very same
> OCaml program is exactly as fast as C.
> 
> 
> http://shootout.alioth.debian.org/u32/performance.php?test=binarytrees
> 
> 
> « Note: these programs are being measured with the default initial
> heap size - the measurements may be very different with a larger
> initial heap size or GC tuning. »
> C version : 12.11 secs
> OCaml version : 47.22 secs
> OCaml version with GC parameters tuned ("interesting alternative"
> section) : 12.67 secs
> 
> 
> 
> 
> Therefore, there is nothing that can be changed to the OCaml
> submission for this benchmark to improve performances, except changing
> the default GC parameters; while this might be a good idea in general,
> changing it only for the sake of shootout-obsessed people is
> ridiculous.

It's in deed an unfair comparison: In C they use the Apache runtime
which provides memory pools. This is something that does not extend to
most real world programs.

Because it's ridiculous anyway: Encode the tree in an array. Not really
idiomatic, but in C they also do not use the idiomatic memory management
(malloc/free).

Gerd
-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

Re: [Caml-list] Is OCaml fast?

[Goswin von Brederlow]

Török Edwin <edwintorok@gmail.com> writes:

> On Mon, 22 Nov 2010 15:36:30 +0100
> bluestorm <bluestorm.dylc@gmail.com> wrote:
>
>> On Mon, Nov 22, 2010 at 3:04 PM, Gerd Stolpmann
>> <info@gerd-stolpmann.de>wrote:
>> 
>> > I think the shootout is not a good data source. There are definitely
>> > some very poor Ocaml results there, so I'd guess the shootout got
>> > recently more attention by enthusiasts of other languages, and the
>> > current Ocaml programs there are not very good. (I remember Ocaml
>> > was #1 at the shootout a few years ago, faster than C.) So maybe a
>> > good opportunity to post better Ocaml solutions there?
>> >
>> 
>> As Sylvain noticed, some (in not most) of the OCaml poor performances
>> in the shootout are actually not due to bad OCaml programs, but to
>> arbitrary restrictions in the shootout rules. For example, one of the
>> bad-performing benchmark for OCaml is the binary-tree benchmark,
>> where it is nearly four times slower than C, but on closer inspection
>> you discover that this is due to the arbitrary choice to forbid any
>> change of the GC parameters. With appropriate GC parameters, the very
>> same OCaml program is exactly as fast as C.
>> 
>> http://shootout.alioth.debian.org/u32/performance.php?test=binarytrees
>> 
>> « Note: these programs are being measured with *the default initial
>> heap size* - the measurements may be very different with a larger
>> initial heap size or GC tuning. »
>> C version : 12.11 secs
>> OCaml version : 47.22 secs
>> OCaml version with GC parameters tuned ("interesting alternative"
>> section) : 12.67 secs
>> 
>> 
>> Therefore, there is nothing that can be changed to the OCaml
>> submission for this benchmark to improve performances, except
>> changing the default GC parameters; while this might be a good idea
>> in general, changing it only for the sake of shootout-obsessed people
>> is ridiculous.
>
> Isn't it possible for the GC to realise its doing too many collections
> and increase the minor heap size on its own?

Maybe. The GC would have to see that with a larger minor heap it has
sufficient space so that most values die before the collection. It could
see that most values die after 1 major sweep and increasing the minor
heap would likely keep them out of the major heap alltogether.

> Or isn't it possible to determine at compile time an approximation for
> a good heap size, and use that?

Usualy that is dependent on the input.

MfG
        Goswin

Re: [Caml-list] Is OCaml fast?

[Fabrice Le Fessant]

On Mon, Nov 22, 2010 at 3:04 PM, Gerd Stolpmann <info@gerd-stolpmann.de> wrote:
> So maybe a good opportunity to post better Ocaml solutions there?

I spent some time improving OCaml solutions, and most of the time, my
solutions were refused: the organizers don't let you unroll loops, fix
GC parameters, etc. One strength of OCaml is that it runs fairly fast
immediatly, but you can make it run even faster with some tuning,
something that you cannot do with many other languages. Such tuning is
not allowed there, so, it is not a good place for OCaml hackers. If
the organizers had been consistent, they would have forbidden Haskell
programmers from adding strictness in their programs, but then, there
programs would have been 2x slower than OCaml programs.

Well, actually, things might improve in the future, if we manage to
improve the performance of the compiler itself, and have more
"official" libraries. For example, most parallel solutions in OCaml
currently use fork() and sockets for communication. Using an official
shared-memory library, we could directly improve the performances of
many OCaml solutions.

--Fabrice

Re: [Caml-list] Is OCaml fast?

[Fabrice Le Fessant]

2010/11/22 Török Edwin <edwintorok@gmail.com>:
> Isn't it possible for the GC to realise its doing too many collections
> and increase the minor heap size on its own?

Indeed, it could notice that a lot of data is being moved to the major
heap, and double its size in consequence, until a maximum limit is
reached.

 The problem is that it is the kind of things that are application
dependent, and should be put in the program itself (the program would
have a trigger on each minor heap collection, and, depending on the
moved bytes, would increase the size of the minor heap). The problem
is that the Shootout does not allow that, so the winner is the
language whose runtime allocates the most memory at the beginnning...

--Fabrice

Re: [Caml-list] Is OCaml fast?

[Oliver Bandel]

Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:
[...]
> (I remember Ocaml was #1
> at the shootout a few years ago, faster than C.) So maybe a good
> opportunity to post better Ocaml solutions there?
[...]

Yes I also remember that.
I hope that the new OCaml compilers did not
make OCaml lessperformance by enhancing other features.

And I don't realy think so.

But were the old code-snippets emoved, or what was going on,
that OCaml degraded that much?

 From my experience - normally using ocamlc for most of my OCaml
programs - it's even fast enough with this bytecode.
If it will be not fast enough, I have an option as a joker:
compiling to native code. :)

Doing this is rather psychologically, as I like to have a potential
for enhancement. :)

If you really always need the best performance, of course compiling
to native code will be best choice.
But at lest for my work I have not found out really bottlenecks,
and always was faster than doing things in Perl or Python or so.

Learning OCaml IMHO does make sense in any case. :)

But be aware one big disadvantage: if you have learned Ocaml,
this may absolutely kill your motivation in learning other languages  
afterwards.So,if you need some other languages for your job, learn  
them first, and learn OCaml after this, for your pleasure.

But you may never have fun at work then, until your boss allows you to  
use OCaml.

And this is not a joke, I mean it as I wrote it!

Ciao,
    Oliver

Re: [Caml-list] Is OCaml fast?

[David Rajchenbach-Teller]

I can confirm that old code-snippets were removed (and that both faster solutions and environment variable tweaks were rejected).

On Nov 22, 2010, at 6:02 PM, Oliver Bandel wrote:

> Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:
> [...]
>> (I remember Ocaml was #1
>> at the shootout a few years ago, faster than C.) So maybe a good
>> opportunity to post better Ocaml solutions there?
> [...]
> 
> Yes I also remember that.
> I hope that the new OCaml compilers did not
> make OCaml lessperformance by enhancing other features.
> 
> And I don't realy think so.
> 
> But were the old code-snippets emoved, or what was going on,
> that OCaml degraded that much?
> 

Re: [Caml-list] Is OCaml fast?

[Philippe Strauss]

answer to the original poster:

ocaml is surprisingly fast for such a high level language, i like this  
insight/comparison:

http://youinfinitesnake.blogspot.com/2010/09/programming-language-wars-movie.html


(even if the benchmarks can be improved reading others posts of this  
thread)

Re: [Caml-list] Is OCaml fast?

[Oliver Bandel]

...hmhhh..

...looks like they are biased...


.... not that we are not ;)

...but... as the GC-stuff is available FROM WITHING the language,
in the standard-lib, this is nothing added on later.


And I think it should also be allowed to be used.

To reject environment variables, I can see as acceptable in this case,
but rejecting the GC-stuff does not make sense, because, as just  
mentioned, it is avalable by the programmer from within the code.

What about compiling parameters?
I mean: in C you can use -O for optimization.
This should also be forbidden then.... Is it?

There are so much possibilities to influence the results,
that blocking Gc-module is idiotic, IMHO.


Ciao,
   Oliver


P.S.:
I looked at one of the C-makefiles:

usr/bin/gcc -pipe -Wall -O3 -fomit-frame-pointer -march=native  
-fopenmp -D_FILE_OFFSET_BITS=64 -I/usr/include/apr-1.0 -lapr-1 -lgomp  
binarytrees.gcc-7.c -o binarytrees.gcc-7.gcc_run
rm binarytrees.gcc-7.c


So, -O3 is allowed.
AFAIK with O3 and higher, inline does work.
__inline__ must be forbidden as well as -O3

Optimization should be switched off completely, if
OCaml's optimizations are also not allowed.





Zitat von "David Rajchenbach-Teller" <David.Teller@univ-orleans.fr>:

> I can confirm that old code-snippets were removed (and that both  
> faster solutions and environment variable tweaks were rejected).
>
> On Nov 22, 2010, at 6:02 PM, Oliver Bandel wrote:
>
>> Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:
>> [...]
>>> (I remember Ocaml was #1
>>> at the shootout a few years ago, faster than C.) So maybe a good
>>> opportunity to post better Ocaml solutions there?
>> [...]
>>
>> Yes I also remember that.
>> I hope that the new OCaml compilers did not
>> make OCaml lessperformance by enhancing other features.
>>
>> And I don't realy think so.
>>
>> But were the old code-snippets emoved, or what was going on,
>> that OCaml degraded that much?
>>
>
>

Re: [Caml-list] Is OCaml fast?

[David Rajchenbach-Teller]

Note: I'm not saying that they are biased. It's quite possible they did the same thing for other languages, too, I didn't take the time to check.

On Nov 22, 2010, at 6:23 PM, Oliver Bandel wrote:

> 
> ...hmhhh..
> 
> ...looks like they are biased...
> 
> 
> .... not that we are not ;)

Re: [Caml-list] Is OCaml fast?

[Oliver Bandel]

Hi,

Zitat von "Thanassis Tsiodras" <ttsiodras@gmail.com>:

> I apologize beforehand if this is not the forum to ask.
>
> I am on the fence about whether to learn OCaml or not, and while
> reading an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw
> that OCaml was praised for the speed of the executables it generates -
> and was referred to, speed-wise, as "second to none", except C and
> C++.
[...]


I remember one argument on that topic, many years ago,
maybe rom Xavier leoy or other core developers, and it goes like that:

Benchmarks on such small problem solving is ONE comparision, what  
makes sense, but it must be added by comparisions of "optimizations"  
of the code in large applications.

The lower the level of the öanguage the harder it is to keep track of
possible optimizations on a very big program.

But the higher the language level is, the easier it is to
keep that in mind.

So: when optmizing small snippets of code this helps optimizing  
performance in maybe critical sections, but to get an overview on all  
interdependencies is close to unmanagable with lower leveled languages.



I have not seen a comparison of performance in this repect,
but I also would think higher level languages might be a better choice here.


Any comparisions on that scale of programming, which you know of?

Then please throw in a link here.

Thanks.

Ciao,
    Oliver

Re: [Caml-list] Is OCaml fast?

[Török Edwin]

On Mon, 22 Nov 2010 17:46:49 +0100
Fabrice Le Fessant <fabrice@lefessant.net> wrote:

> 2010/11/22 Török Edwin <edwintorok@gmail.com>:
> > Isn't it possible for the GC to realise its doing too many
> > collections and increase the minor heap size on its own?
> 
> Indeed, it could notice that a lot of data is being moved to the major
> heap, and double its size in consequence, until a maximum limit is
> reached.

I did some benchmarks on 2 CPUs, and here are the wall clock times
for GC minor heap size:

"minorheap","      phenomII",,,"               core2",,,"cycle diff %"
,"              time",,"      cycles","time",,"      cycles",
32,"           17.76","17.74",56.8," 48.60","48.86",58.48,2.96
64,"           16.81","16.80",53.78,"44.79","44.79",53.75,-0.06
128,"          15.41","15.38",49.26,"41.38","40.76",49.28,0.04
256,"          14.39","14.35",45.98,"39.75","38.98",47.24,2.74
512,"          12.97","13.15",41.79,"35.75","35.59",42.8,2.42
1024,"         12.89","12.94",41.33,"33.57","33.13",40.02,-3.17
2048,"         11.05","11.09",35.42,"29.26","29.12",35.03,-1.1
4096,"         9.79"," 9.81",31.36,"26.21","25.96",31.3,-0.19
8192,"         8.71"," 8.85",28.1," 23.36","23.34",28.02,-0.28
16384,"        7.89"," 7.86",25.2," 21.41","21.54",25.77,2.26
32768,"        7.55"," 7.55",24.16,"20.74","20.88",24.97,3.35

(minorheap is in KWords, time is in seconds, cycles is divided by 10^9)

Increasing minor heap beyond that yielded no improvement (number of
minor heap collections stayed the same).

phenomII has 64 Kb L1 data cache, 512Kb L2 cache, 6144Kb L3 cache
(shared), runs at 3.2Ghz. That would be 516k words if only 1 core used.

core2  has 32 Kb L1 data cache, 4MB L2 cache, runs at 1.2Ghz.
That would be 840k words if only 1 core used.

Both used exact same binaries on 64-bit Linux, ocaml 3.11.2.

Despite the difference in CPUs and heap sizes the number of CPU cycles
spent for a given size of the minor heap is about the same (within 3%
difference).

Not sure what the max should be for the minor heap increase, but based
on this benchmark increasing size of minor heap never slows down the
program. Even when size of minor heap exceeds what fits in the cache.
I guess there is another microbenchmark that can show the opposite
though.

Is there some real world benchmark for OCaml's GC that can be used
to get some better values for minor heap size based on CPU's L2/L3 size
(as suggested in the 'Optimizing garbage collection' thread)?


> 
>  The problem is that it is the kind of things that are application
> dependent,

Yes, maybe optimal size of minor heap doesn't depend on CPU's cache
size but on the application, in which case a heuristic for
increasing/decreasing the heap size may be better?

> and should be put in the program itself (the program would
> have a trigger on each minor heap collection, and, depending on the
> moved bytes, would increase the size of the minor heap). The problem
> is that the Shootout does not allow that, so the winner is the
> language whose runtime allocates the most memory at the beginnning...

If tuning minor heap can double performance of the program, then the GC
should have some heuristics to tune it automatically. Sure applications
which are not happy with that tuning should tune it themselves.

Best regards,
--Edwin

Re: [Caml-list] Is OCaml fast?

[Eray Ozkural]

[-- Attachment #1: Type: text/plain, Size: 1346 bytes --]

It's always funny to look at benchmarks made by amateurs. We get it all the
time in parallel computing, when some sharp programmer decides to implement
a parallel algorithm in Java. :) I keep asking myself questions like do you
also program ciphers in Flash? Graph algorithms in PHP?

In my experience, ocaml is pretty fast, quite comparable to C++, for
computation-intensive tasks and of course excellent at traditional
functional programming tasks. Since I use both kinds of algorithms in my
research, ocaml is better than C++ for me (despite the great boost libs),
except for shared memory parallelism, which is still non-existent in ocaml.

Try implementing something that has a large time/memory complexity like a
data mining algorithm, and you'll see the difference from Haskell/Java/Blah.
Just run the algorithms in Weka, the famous data mining suite written in
Java, and compare them to any sensible implementation of the same
algorithms. Measure the difference, it's unbelievable how awful Java is at
any kind of serious computing! I think some conservative humans from the
future sent the designers of Java back in time to stop the singularity.

Cheers,

-- 
Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo.com/group/ai-philosophy
http://myspace.com/arizanesil http://myspace.com/malfunct

[-- Attachment #2: Type: text/html, Size: 1639 bytes --]

Re: [Caml-list] Is OCaml fast?

[Jeff Schultz]

On Mon, Nov 22, 2010 at 06:23:38PM +0100, Oliver Bandel wrote:
> ...looks like they are biased...

> .... not that we are not ;)

> To reject environment variables, I can see as acceptable in this case,
> but rejecting the GC-stuff does not make sense, because, as just mentioned, 
> it is avalable by the programmer from within the code.

I haven't looked at the "benchmark game" or its predecessor for many
years, but all this really says is that the programs it measures are
sufficiently unlike real applications that the OCaml developers
haven't been tempted to tune the GC defaults for them.

The conclusion to draw would be that the shootout results are a
poor, and probably misleading, evidence base for selecting a
programming language for execution time performance.  Measured
programming language implementation performance gets faster or
slower over the years due to its own rule changes and language and
library implementation changes, but I doubt that this conclusion
changes much ;-)


The one thing the shootout programs are useful for is to get a feel
for a variety of different languages' expressivity for a common set
of "small" problems.


    Jeff Schultz

Re: Is OCaml fast?

[Isaac Gouy]

Thanassis Tsiodras <ttsiodras <at> gmail.com> writes:

-snip-
> Is it just hype, then? Or am I missing something?



Note the "Why Ocaml?" date - December 2002. 

Re: Is OCaml fast?

[Isaac Gouy]

David Rajchenbach-Teller <David.Teller <at> univ-orleans.fr> writes:


> I can confirm that old code-snippets were removed (and that both faster
solutions and environment
> variable tweaks were rejected).


Even back in 2001, Doug Bagley had noted all the things that were
wrong with the tasks on his "The Great Computer Language Shootout".

During autumn 2004, on Brent Fulgham's website, new tasks were added
and old Doug Bagley tasks removed.

By 2005 only 2 tasks remained from that old Doug Bagley website.

By August 2008 none of those old Doug Bagley tasks were measured for
the current benchmarks game. 

Re: Is OCaml fast?

[Isaac Gouy]

bluestorm <bluestorm.dylc <at> gmail.com> writes:

-snip-
> With appropriate GC parameters, the very same OCaml program is exactly as fast
as C.
> 
> 
> http://shootout.alioth.debian.org/u32/performance.php?test=binarytrees
> 
> « Note: these programs are being measured with the default initial heap size -
the measurements may be very different with a larger initial heap size or GC
tuning. »
> 
> 
> C version : 12.11 secs
> OCaml version : 47.22 secs
> OCaml version with GC parameters tuned ("interesting alternative" section) :
12.67 secs



And of course you know because that GC tuned OCaml program is shown on the
benchmarks game website ;-) 

http://shootout.alioth.debian.org/u32/program.php?test=binarytrees&lang=ocaml&id=1

Re: Is OCaml fast?

[Isaac Gouy]

Gerd Stolpmann <info <at> gerd-stolpmann.de> writes:

-snip-
> I think the shootout is not a good data source. There are definitely
> some very poor Ocaml results there, so I'd guess the shootout got
> recently more attention by enthusiasts of other languages, and the
> current Ocaml programs there are not very good. (I remember Ocaml was #1
> at the shootout a few years ago, faster than C.) So maybe a good
> opportunity to post better Ocaml solutions there?


Even back in 2001, on Doug Bagley's website called "The Great Computer Language
Shootout"  Ocaml was not #1-

http://web.archive.org/web/20010602051335/www.bagley.org/~doug/shootout/craps.shtml

Doug Bagley stopped updating his website in Fall 2001.


Even back in 2004, when Brent Fulgham started his website using the same
programs, OCaml was not #1

http://web.archive.org/web/20040717184911/shootout.alioth.debian.org/craps.php

Re: Is OCaml fast?

[Isaac Gouy]

Dario Teixeira <darioteixeira <at> yahoo.com> writes:
-snip-
> There's lies, damn lies, and shootout statistics. 
-snip-

After all, facts are facts,
and although we may quote one to another with a chuckle
the words of the Wise Statesman, 'Lies--damned lies--and statistics,'
still there are some easy figures the simplest must understand,
and the astutest cannot wriggle out of.

Leonard Henry Courtney, 1895

:-) 

Re: [Caml-list] Re: Is OCaml fast?

[Fabrice Le Fessant]

Maybe you should read "Tainted Truth: The Manipulation of Fact In
America" by Cynthia Crossen ?

--Fabrice

Isaac Gouy wrote, On 11/23/2010 03:20 AM:
> Dario Teixeira <darioteixeira <at> yahoo.com> writes:
> -snip-
>> There's lies, damn lies, and shootout statistics. 
> -snip-
> 
> After all, facts are facts,
> and although we may quote one to another with a chuckle
> the words of the Wise Statesman, 'Lies--damned lies--and statistics,'
> still there are some easy figures the simplest must understand,
> and the astutest cannot wriggle out of.
> 
> Leonard Henry Courtney, 1895

Re: [Caml-list] Re: Is OCaml fast?

[Christophe TROESTLER]

On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
>
> > C version : 12.11 secs
> > OCaml version : 47.22 secs
> > OCaml version with GC parameters tuned ("interesting alternative" 
>  section) : 12.67 secs
>
> And of course you know because that GC tuned OCaml program is shown 
>  on the
> benchmarks game website ;-)
> http://shootout.alioth.debian.org/u32/program.php?test=binarytrees& 
> lang=ocaml&id=1

Since you are here, please explain why C can use memory pools and vec 
tor instructions but tuning the GC of OCaml ― although it is part of  
the standard library ― is considered an “alternative”.

Best,
C.

RE: [Caml-list] Re: Is OCaml fast?

[Jon Harrop]

And almost all of the "Haskell" solutions (reverse-complement,
spectral-norm, Mandelbrot, n-body, fannkuch-redux, k-nucleotide, regex-dna)
abuse GHC's FFI in order to work around Haskell.

The k-nucleotide benchmark in Haskell even uses malloc! Ketil Malde crafted
a much better solution but noted:

  "This is considered cheating, since it is the easy and natural way to do
it." -
http://www.haskell.org/haskellwiki/Shootout/Knucleotide

Cheers,
Jon.

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Christophe TROESTLER
> Sent: 23 November 2010 10:38
> To: igouy2@yahoo.com
> Cc: caml-list@inria.fr
> Subject: Re: [Caml-list] Re: Is OCaml fast?
> 
> On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
> >
> > > C version : 12.11 secs
> > > OCaml version : 47.22 secs
> > > OCaml version with GC parameters tuned ("interesting alternative"
> >  section) : 12.67 secs
> >
> > And of course you know because that GC tuned OCaml program is shown
> >  on the
> > benchmarks game website ;-)
> > http://shootout.alioth.debian.org/u32/program.php?test=binarytrees&
> > lang=ocaml&id=1
> 
> Since you are here, please explain why C can use memory pools and vec
> tor instructions but tuning the GC of OCaml ― although it is part of
> the standard library ― is considered an "alternative".
> 
> Best,
> C.
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

RE: [Caml-list] Re: Is OCaml fast?

[Jon Harrop]

Note that the regex-dna solution for Haskell tweaks its GC parameters via
the -H command-line parameter:

http://shootout.alioth.debian.org/u64q/program.php?test=regexdna&lang=ghc&id
=2

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Christophe TROESTLER
> Sent: 23 November 2010 10:38
> To: igouy2@yahoo.com
> Cc: caml-list@inria.fr
> Subject: Re: [Caml-list] Re: Is OCaml fast?
> 
> On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
> >
> > > C version : 12.11 secs
> > > OCaml version : 47.22 secs
> > > OCaml version with GC parameters tuned ("interesting alternative"
> >  section) : 12.67 secs
> >
> > And of course you know because that GC tuned OCaml program is shown
> >  on the
> > benchmarks game website ;-)
> > http://shootout.alioth.debian.org/u32/program.php?test=binarytrees&
> > lang=ocaml&id=1
> 
> Since you are here, please explain why C can use memory pools and vec
> tor instructions but tuning the GC of OCaml ― although it is part of
> the standard library ― is considered an "alternative".
> 
> Best,
> C.
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: Is OCaml fast?

[Isaac Gouy]

Christophe TROESTLER <Christophe.Troestler+ocaml <at> umh.ac.be> writes:

> 
> On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
> >
> > > C version : 12.11 secs
> > > OCaml version : 47.22 secs
> > > OCaml version with GC parameters tuned ("interesting alternative" 
> >  section) : 12.67 secs
> >
> > And of course you know because that GC tuned OCaml program is shown 
> >  on the
> > benchmarks game website 
> > http://shootout.alioth.debian.org/u32/program.php?test=binarytrees& 
> > lang=ocaml&id=1
> 
> Since you are here, please explain why C can use memory pools and vec 
> tor instructions but tuning the GC of OCaml ― although it is part of  
> the standard library ― is considered an “alternative”.


You seem to be suggesting that "tuning the GC" is considered "alternative" only
for OCaml programs.

You seem to be suggesting that "tuning the GC" is considered "alternative" for
every task.

Neither is true.

You seem to be suggesting some kind of equivalence between vector instructions
and "tuning the GC".
You haven't said why they should be considered equivalent.

Nor have you said why you think C should not be allowed to use memory pools.

Re: Is OCaml fast?

[Isaac Gouy]

Is that book the source for the quotation 'Lies--damned lies--and statistics'?


Fabrice Le Fessant <fabrice.le_fessant <at> inria.fr> writes:
> Maybe you should read "Tainted Truth: The Manipulation of Fact In
> America" by Cynthia Crossen ?

Re: Is OCaml fast?

[Isaac Gouy]

Jon Harrop <jonathandeanharrop <at> googlemail.com> writes:

> 
> Note that the regex-dna solution for Haskell tweaks its GC parameters via
> the -H command-line parameter:


Note that there is no restriction on "tuning the GC" for regex-dna.

Note that there is no restriction on "tuning the GC" for any task except
binary-trees.

Re: Is OCaml fast?

[Isaac Gouy]

Jon Harrop <jonathandeanharrop <at> googlemail.com> writes:

> Ketil Malde crafted
> a much better solution but noted:
> 
>   "This is considered cheating, since it is the easy and natural way to do
> it." -
> http://www.haskell.org/haskellwiki/Shootout/Knucleotide


Not even cheating - just an answer to a different question.

Re: [Caml-list] Re: Is OCaml fast?

[Török Edwin]

On Tue, 23 Nov 2010 18:03:10 +0000 (UTC)
Isaac Gouy <igouy2@yahoo.com> wrote:

> Jon Harrop <jonathandeanharrop <at> googlemail.com> writes:
> 
> > 
> > Note that the regex-dna solution for Haskell tweaks its GC
> > parameters via the -H command-line parameter:
> 
> 
> Note that there is no restriction on "tuning the GC" for regex-dna.
> 
> Note that there is no restriction on "tuning the GC" for any task
> except binary-trees.

Sounds good. Then Ocaml could still win if it performs well on the
other benchmarks.

Best regards,
--Edwin

Re: [Caml-list] Re: Is OCaml fast?

[Gerd Stolpmann]

Am Dienstag, den 23.11.2010, 17:53 +0000 schrieb Isaac Gouy:
> Christophe TROESTLER <Christophe.Troestler+ocaml <at> umh.ac.be> writes:
> 
> > 
> > On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
> > >
> > > > C version : 12.11 secs
> > > > OCaml version : 47.22 secs
> > > > OCaml version with GC parameters tuned ("interesting alternative" 
> > >  section) : 12.67 secs
> > >
> > > And of course you know because that GC tuned OCaml program is shown 
> > >  on the
> > > benchmarks game website 
> > > http://shootout.alioth.debian.org/u32/program.php?test=binarytrees& 
> > > lang=ocaml&id=1
> > 
> > Since you are here, please explain why C can use memory pools and vec 
> > tor instructions but tuning the GC of OCaml ― although it is part of  
> > the standard library ― is considered an “alternative”.
> 
> 
> You seem to be suggesting that "tuning the GC" is considered "alternative" only
> for OCaml programs.
> 
> You seem to be suggesting that "tuning the GC" is considered "alternative" for
> every task.
> 
> Neither is true.
> 
> You seem to be suggesting some kind of equivalence between vector instructions
> and "tuning the GC".
> You haven't said why they should be considered equivalent.
> 
> Nor have you said why you think C should not be allowed to use memory pools.

Quite easy: because you are comparing results that cannot be compared.
The reader of this benchmark (binary trees) might have the impression
that C is generally that fast - however, this would be no longer true if
these binary trees were used as library in a bigger program where using
memory pools is inappropriate, e.g. because the data managed by the
binary trees has an unpredictable lifetime.

I do not say that it is complete nonsense to do this comparison, but
only that it is more specific than a reader would assume. The innocent
reader expects a comparison of binary tree performance, not of methods
of managing memory (and this is it finally). The true name of this test
should be "manage_many_small_memory_cells_where_pools_are_allowed". (It
would be actually interesting to compare various versions of this test
with different memory management methods.)

Gerd

> 
> 
> 
> 
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
> 


-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

[Caml-list] Re: Is OCaml fast?

[Mark Diekhans]

The source of this pharse is unknown, however Mark Twain is credited with
making it well known:

   "There are three kinds of lies: lies, damned lies and statistics."


Isaac Gouy <igouy2@yahoo.com> writes:
> Is that book the source for the quotation 'Lies--damned lies--and statistics'?
> 
> 
> Fabrice Le Fessant <fabrice.le_fessant <at> inria.fr> writes:
> > Maybe you should read "Tainted Truth: The Manipulation of Fact In
> > America" by Cynthia Crossen ?
> 
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: Is OCaml fast?

[Isaac Gouy]

Mark Diekhans <markd <at> kermodei.com> writes:

> 
> 
> The source of this pharse is unknown, however Mark Twain is credited with
> making it well known:
> 
>    "There are three kinds of lies: lies, damned lies and statistics."


http://www.york.ac.uk/depts/maths/histstat/lies.htm

Re: Is OCaml fast?

[Isaac Gouy]

Török Edwin <edwintorok <at> gmail.com> writes:

-snip- 
> Sounds good. Then Ocaml could still win if it performs well on the
> other benchmarks.


The main benchmarks game summary is median and quartiles so one measurement
doesn't have much influence.

There are several tasks which have worse performing OCaml programs than
binary-trees.

Re: Is OCaml fast?

[Isaac Gouy]

Gerd Stolpmann <info <at> gerd-stolpmann.de> writes:

-snip-
> I do not say that it is complete nonsense to do this comparison, but
> only that it is more specific than a reader would assume.

A reader's wrong assumptions are their own responsibility:

http://shootou.alioth.debian.org/flawed-benchmarks.php


> The innocent reader expects a comparison of binary tree performance,
> not of methods of managing memory (and this is it finally).

Perhaps rather than "innocent reader" you mean careless reader who didn't bother
to read what the programs should do?

http://shootout.alioth.debian.org/u32q/benchmark.php?test=binarytrees&lang=all#about


> The true name of this test should be 
> "manage_many_small_memory_cells_where_pools_are_allowed". 

"binary-trees benchmark : Allocate and deallocate many many binary trees"



> (It would be actually interesting to compare various versions of this test
> with different memory management methods.)

So do that comparison and publish the results.

Re: [Caml-list] Re: Is OCaml fast?

[Gerd Stolpmann]

Am Dienstag, den 23.11.2010, 20:28 +0000 schrieb Isaac Gouy:

> > (It would be actually interesting to compare various versions of this test
> > with different memory management methods.)
> 
> So do that comparison and publish the results.

Please don't tell me what I am supposed to do. I'm not a troll like
others.

Gerd
-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

Re: [Caml-list] Re: Is OCaml fast?

[Christophe TROESTLER]

On Tue, 23 Nov 2010 17:53:14 +0000, Isaac Gouy wrote:
> 
> Christophe TROESTLER writes:
> 
> > Since you are here, please explain why C can use memory pools and vec 
> > tor instructions but tuning the GC of OCaml ― although it is part of  
> > the standard library ― is considered an “alternative”.
> 
> [...] You seem to be suggesting some kind of equivalence between vector
> instructions and "tuning the GC".  You haven't said why they should
> be considered equivalent.

I did not say they are equivalent.  It is a matter of what is allowed
(or not) and for what reason.  The question is why is C allowed to use
an external library for managing its memory but — for this specific
benchmark which is precisely about managing memory — OCaml is not
authorized to make use of its very own library!

On Tue, 23 Nov 2010 18:03:10 +0000, Isaac Gouy wrote:
> 
> Note that there is no restriction on "tuning the GC" for regex-dna.
> 
> Note that there is no restriction on "tuning the GC" for any task except
> binary-trees.

Note that you did not answer my question.

Re: Is OCaml fast?

[Isaac Gouy]

Gerd Stolpmann <info <at> gerd-stolpmann.de> writes:

> > > (It would be actually interesting to compare various versions of this test
> > > with different memory management methods.)
> > 
> > So do that comparison and publish the results.
> 
> Please don't tell me what I am supposed to do. I'm not a troll like
> others.


The tone of my comment was not as harsh as you heard.

RE: [Caml-list] Re: Is OCaml fast?

[Jon Harrop]

Yes, an answer to a better question.

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Isaac Gouy
> Sent: 23 November 2010 18:07
> To: caml-list@inria.fr
> Subject: [Caml-list] Re: Is OCaml fast?
> 
> Jon Harrop <jonathandeanharrop <at> googlemail.com> writes:
> 
> > Ketil Malde crafted
> > a much better solution but noted:
> >
> >   "This is considered cheating, since it is the easy and natural way
> to do
> > it." -
> > http://www.haskell.org/haskellwiki/Shootout/Knucleotide
> 
> 
> Not even cheating - just an answer to a different question.
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: Is OCaml fast?

[Isaac Gouy]

Christophe TROESTLER <Christophe.Troestler+ocaml <at> umh.ac.be> writes:

-snip-
> The question is why is C allowed to use
> an external library for managing its memory

I asked why you think C should not be allowed to use memory pools - you haven't
tried to answer that question.

If you think that C should not be allowed to use memory pools because they are
provided in an external library then I disagree - I think it's fine to show use
of popular external libraries.

Of course, a C program that just uses malloc is also shown.


> but — for this specific
> benchmark which is precisely about managing memory — OCaml is not
> authorized to make use of its very own library!

OCaml is authorized to make use of its very own library.

Just like all the other programming language implementations, for binary-trees,
OCaml is not authorized to tune the GC.

Re: [Caml-list] Re: Is OCaml fast?

[Richard Jones]

On Tue, Nov 23, 2010 at 06:03:10PM +0000, Isaac Gouy wrote:
> Jon Harrop <jonathandeanharrop <at> googlemail.com> writes:
> 
> > 
> > Note that the regex-dna solution for Haskell tweaks its GC parameters via
> > the -H command-line parameter:
> 
> 
> Note that there is no restriction on "tuning the GC" for regex-dna.
> 
> Note that there is no restriction on "tuning the GC" for any task except
> binary-trees.

Don't you think this is a pretty ludicrous restriction?

Tuning the GC / adjusting the size of pools or mallocs is essential
for optimizing for modern processors.  I bet the C programs are doing
this, except that it won't obviously be called "tuning the GC"
although it amounts to precisely the same thing.

Rich.

-- 
Richard Jones
Red Hat

Re: [Caml-list] Is OCaml fast?

[oliver]

On Mon, Nov 22, 2010 at 05:55:02AM -0800, Dario Teixeira wrote:
> Hi,
> 
> > I am on the fence about whether to learn OCaml or not, and while
> > reading an article called "Why OCaml"
> > (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html),
> > I saw that OCaml was praised for the speed of the executables it
> > generates - and was referred to, speed-wise, as "second to none",
> > except C and C++.
> 
> Yes, Ocaml is fast (more on that later), but you should consider speed
> as just the icing in the cake.
[...]

But if it were not performant enough, I (and many others too, I think)
would consider it being a nice toy language.

AFAIK in the past, functional langauges were not adapted, because they were
very unperformant - at least this is one reason.
Another reason might be, that the available functional languages in the past
were overloaded with parenthess ;)

Ciao,
   Oliver

Re: Is OCaml fast?

[evil sloot]

Hi,

Regarding alternative solutions to the shootout problems:
The thread ring problem can be easily implemented using Lwt
yielding (pun intended ;)) performance 
comparable to the haskell solution

have fun,

Romain.



      

Re: [Caml-list] Re: Is OCaml fast?

[oliver]

On Tue, Nov 23, 2010 at 02:01:33AM +0000, Isaac Gouy wrote:
> David Rajchenbach-Teller <David.Teller <at> univ-orleans.fr> writes:
> 
> 
> > I can confirm that old code-snippets were removed (and that both faster
> solutions and environment
> > variable tweaks were rejected).
> 
> 
> Even back in 2001, Doug Bagley had noted all the things that were
> wrong with the tasks on his "The Great Computer Language Shootout".


And what was wrong in his eyes?


> 
> During autumn 2004, on Brent Fulgham's website, new tasks were added
> and old Doug Bagley tasks removed.
> 
> By 2005 only 2 tasks remained from that old Doug Bagley website.
> 
> By August 2008 none of those old Doug Bagley tasks were measured for
> the current benchmarks game. 

So, now the comparisions are perfect?
What problems were removed?


Ciao,
   Oliver

RE: [Caml-list] Is OCaml fast?

[Jon Harrop]

Oliver wrote:
> AFAIK in the past, functional langauges were not adapted, because they
> were
> very unperformant - at least this is one reason.
> Another reason might be, that the available functional languages in the
> past
> were overloaded with parenthess ;)

That was also true of early ML implementations. When I was first taught ML at university we used the Cambridge ML interpreter and you actually had to sit there and wait for it to solve the 8-queens problem. At the time, I thought ML was a complete joke and could see no use for it outside its very specific domain of theorem proving and actually really resented being taught it on a general CS course. I know better now though. ;-)

Later language implementations inherited many of these inefficiencies though. Many of the things that can make OCaml and Java slow were inherited from Lisp. They are, in effect, designs that still bear the burden of dynamic typing despite being statically typed.

Cheers,
Jon.

Re: Is OCaml fast?

[Isaac Gouy]

Richard Jones <rich <at> annexia.org> writes:

> > Note that there is no restriction on "tuning the GC" for regex-dna.
> > 
> > Note that there is no restriction on "tuning the GC" for any task except
> > binary-trees.
> 
> Don't you think this is a pretty ludicrous restriction?
> 
> Tuning the GC / adjusting the size of pools or mallocs is essential
> for optimizing for modern processors.  I bet the C programs are doing
> this, except that it won't obviously be called "tuning the GC"
> although it amounts to precisely the same thing.


What you say would be interesting if you'd taken the trouble to look at the C
programs.

Re: Is OCaml fast?

[Isaac Gouy]

 <oliver <at> first.in-berlin.de> writes:
 
> > Even back in 2001, Doug Bagley had noted all the things that were
> > wrong with the tasks on his "The Great Computer Language Shootout".
> 
> And what was wrong in his eyes?

Find out for yourself:

http://web.archive.org/web/20010617014807/www.bagley.org/~doug/shootout/


> So, now the comparisions are perfect?

Has anyone said so?


> What problems were removed?

All of them.

Re: [Caml-list] Is OCaml fast?

[Erik de Castro Lopo]

Jon Harrop wrote:

> Many of the things that can make OCaml and Java slow were inherited
> from Lisp. They are, in effect, designs that still bear the burden
> of dynamic typing despite being statically typed.

I'm curious, what in your opinion are those things?

Erik
-- 
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/

Re: [Caml-list] Re: Is OCaml fast?

[Eray Ozkural]

[-- Attachment #1: Type: text/plain, Size: 3090 bytes --]

Hello,

I think that this benchmark is lacking in the algorithms department. Where
is a dynamic programming problem? A graph algorithm? Anything with
non-trivial time/space complexity? Anything a little more complex than
matrix product?

Also, it's not uncommon to disallow low-level optimizations such as writing
memory allocators and async file access when comparing implementations of an
algorithm, but such restrictions should be carried out uniformly.  In such a
benchmark I would expect each entry to stick to their guns, i.e. use only
the standard libraries and programming styles for instance. Linking in
foreign libraries must most definitely be disallowed. So, if in Java, it's
necessary to call the garbage collector explicitly from time to time, and we
had to do that for a long time, so be it. Or again, if in Java, performance
will suffer unless you only use arrays of integral types, the implementer
may wish to implement as much as is possible with arrays, though I wonder if
it is not better to choose the most natural implementation style for the
particular language. In the case of Java, the claim was that object-oriented
was some kind of a programming-aid that can replace talented programmers :)
It's unfortunate of course that some kinds of optimizations always have to
be made by hand, for instance in functional languages many compilers do not
have deforestation.

Otherwise, of course, any implementation may include a compiler for the
fastest language and present a program in that language, which is not the
objective.

An alternative objective could be to compare the shortest and most
comprehensible, if possible line-to-line compatible implementation of a
given pseudocode in different languages. That would be extremely informative
for serious algorithm researchers! If a computer scientist isn't sure of the
performance of the primitives, he cannot make sure his implementation will
comply with the time-complexity of the given algorithm.

Best Regards,

On Wed, Nov 24, 2010 at 2:23 AM, Isaac Gouy <igouy2@yahoo.com> wrote:

>  <oliver <at> first.in-berlin.de> writes:
>
> > > Even back in 2001, Doug Bagley had noted all the things that were
> > > wrong with the tasks on his "The Great Computer Language Shootout".
> >
> > And what was wrong in his eyes?
>
> Find out for yourself:
>
> http://web.archive.org/web/20010617014807/www.bagley.org/~doug/shootout/<http://web.archive.org/web/20010617014807/www.bagley.org/%7Edoug/shootout/>
>
>
> > So, now the comparisions are perfect?
>
> Has anyone said so?
>
>
> > What problems were removed?
>
> All of them.
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>



-- 
Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo.com/group/ai-philosophy
http://myspace.com/arizanesil http://myspace.com/malfunct

[-- Attachment #2: Type: text/html, Size: 4168 bytes --]

Re: Is OCaml fast?

[Isaac Gouy]

Eray Ozkural <examachine <at> gmail.com> writes:


> Hello, I think that this benchmark is lacking ...

http://shootout.alioth.debian.org/help.php#why


Please make the kind of comparison you think should be done and publish it.

Re: [Caml-list] Re: Is OCaml fast?

[Jeff Meister]

Everyone in this thread is capable of reading your site and has
probably already done so. We know what your rules are for
binary-trees; repeating them does not help. Richard's objection, which
you dismissed out of hand, was that your no-GC-tuning rule is silly in
the light of actual uses of garbage collected programming languages on
modern processors. It makes your results unrealistic, and an
unrealistic benchmark is misleading, or at best merely useless. You
are free to tersely reject our constructive criticism, but the only
meaningful consequence will be that OCaml users consider the shootout
untrustworthy and completely ignore its results... what good are the
"language comparisons" your project makes if the communities behind
those languages don't support your benchmarking methods?

On Tue, Nov 23, 2010 at 6:13 PM, Isaac Gouy <igouy2@yahoo.com> wrote:
> Eray Ozkural <examachine <at> gmail.com> writes:
>
>
>> Hello, I think that this benchmark is lacking ...
>
> http://shootout.alioth.debian.org/help.php#why
>
>
> Please make the kind of comparison you think should be done and publish it.
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>

RE: [Caml-list] Re: Is OCaml fast?

[Andrew]

> ::Jeff Meister

>Everyone in this thread is capable of reading your site and has
>probably already done so. We know what your rules are for
>binary-trees; repeating them does not help. Richard's objection, which
>you dismissed out of hand, was that your no-GC-tuning rule is silly in
>the light of actual uses of garbage collected programming languages on
>modern processors. It makes your results unrealistic, and an
>unrealistic benchmark is misleading, or at best merely useless. You
>are free to tersely reject our constructive criticism, but the only
>meaningful consequence will be that OCaml users consider the shootout
>untrustworthy and completely ignore its results... what good are the
>"language comparisons" your project makes if the communities behind
>those languages don't support your benchmarking methods?

+1. Seriously, Isaac, try to calm down, everything is fine. You might want
to read what others write, I have the feeling that many people were making
valid points, whereas you have mostly been turning down any objection by
pointing people to the same webpage and again.

Andrew.

Re: Is OCaml fast?

[Isaac Gouy]

Jeff Meister <nanaki <at> gmail.com> writes:


> We know what your rules are for
> binary-trees; repeating them does not help. 

When Christophe TROESTLER wrongly states - "OCaml is not authorized to make use
of its very own library!" - he shows that those rules are not known.


> Richard's objection, which you dismissed out of hand, was that your 
> no-GC-tuning rule is silly in the light of actual uses of garbage collected
> programming languages on modern processors. 

When said Richard opines about programs he apparently hasn't bothered to read, I
take that as a sign his opinions might not be based on anything solid.


> It makes your results unrealistic, and an
> unrealistic benchmark is misleading, or at best merely useless.

You should think that benchmarks (not just these) are unrealistic - your
application is the ultimate benchmark.

Useless? Wouldn't that depend on the objectives? The post you replied to
linked-to a 3 line statement of objectives - did you read it? 


> You are free to tersely reject our constructive criticism

Do you really think saying something is "ludicrous" or "silly" is constructive
criticism? :-)

Re: [Caml-list] Re: Is OCaml fast?

[David Rajchenbach-Teller]

Submitted this a few years ago (just with streams, not Lwt), and the solution was rejected.


On Nov 24, 2010, at 12:21 AM, evil sloot wrote:

> Hi,
> 
> Regarding alternative solutions to the shootout problems:
> The thread ring problem can be easily implemented using Lwt
> yielding (pun intended ;)) performance 
> comparable to the haskell solution
> 
> have fun,
> 
> Romain.
> 
> 
> 
> 
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] Re: Is OCaml fast?

[David Rajchenbach-Teller]

[-- Attachment #1: Type: text/plain, Size: 2848 bytes --]

Maybe the solution is to get one of our numbers among the maintainers of the shootout. This would guarantee, if not objectivity, then at least informed choices wrt OCaml.

On Nov 24, 2010, at 2:36 AM, Eray Ozkural wrote:

> Hello,
> 
> I think that this benchmark is lacking in the algorithms department. Where is a dynamic programming problem? A graph algorithm? Anything with non-trivial time/space complexity? Anything a little more complex than matrix product?
> 
> Also, it's not uncommon to disallow low-level optimizations such as writing memory allocators and async file access when comparing implementations of an algorithm, but such restrictions should be carried out uniformly.  In such a benchmark I would expect each entry to stick to their guns, i.e. use only the standard libraries and programming styles for instance. Linking in foreign libraries must most definitely be disallowed. So, if in Java, it's necessary to call the garbage collector explicitly from time to time, and we had to do that for a long time, so be it. Or again, if in Java, performance will suffer unless you only use arrays of integral types, the implementer may wish to implement as much as is possible with arrays, though I wonder if it is not better to choose the most natural implementation style for the particular language. In the case of Java, the claim was that object-oriented was some kind of a programming-aid that can replace talented programmers :) It's unfortunate of course that some kinds of optimizations always have to be made by hand, for instance in functional languages many compilers do not have deforestation.
> 
> Otherwise, of course, any implementation may include a compiler for the fastest language and present a program in that language, which is not the objective.
> 
> An alternative objective could be to compare the shortest and most comprehensible, if possible line-to-line compatible implementation of a given pseudocode in different languages. That would be extremely informative for serious algorithm researchers! If a computer scientist isn't sure of the performance of the primitives, he cannot make sure his implementation will comply with the time-complexity of the given algorithm.
> 
> Best Regards,
> 
> On Wed, Nov 24, 2010 at 2:23 AM, Isaac Gouy <igouy2@yahoo.com> wrote:
> 
> 
> 
> -- 
> Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
> http://groups.yahoo.com/group/ai-philosophy
> http://myspace.com/arizanesil http://myspace.com/malfunct
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs


[-- Attachment #2: Type: text/html, Size: 3435 bytes --]

Re: Is OCaml fast?

[Isaac Gouy]

Andrew <newsgroups.fr <at> gmail.com> writes:

> +1. Seriously, Isaac, try to calm down, everything is fine. You might want
> to read what others write, I have the feeling that many people were making
> valid points, whereas you have mostly been turning down any objection by
> pointing people to the same webpage and again.

Seriously, Andrew, I have been calm.

You might want to read what others write.

I have the feeling that a list of those "valid points" would be a better basis
for discussion than your feeling.

A quick count shows 8 URIs all different - so you are plainly wrong, I have not
been "pointing people to the same webpage and again".

Re: Is OCaml fast?

[Isaac Gouy]

David Rajchenbach-Teller <David.Teller <at> univ-orleans.fr> writes:


> Maybe the solution is to get one of our numbers among the maintainers of the
shootout. This would guarantee, if not objectivity, then at least informed
choices wrt OCaml.

Presumably you'd need all of the maintainers to be OCaml certified to guarantee
"objectivity" :-)

Re: [Caml-list] Re: Is OCaml fast?

[Christophe Troestler]

On Wed, 24 Nov 2010 06:50:15 +0000, Isaac Gouy wrote:
> 
> Jeff Meister <nanaki <at> gmail.com> writes:
> 
> > We know what your rules are for binary-trees; repeating them does
> > not help.
> 
> When Christophe TROESTLER wrongly states - "OCaml is not authorized
> to make use of its very own library!" - he shows that those rules
> are not known.

Isolating a sentence gives a misleading idea of what I said.  Must I
really repeat that I was asking WHY is C allowed to use an external
library to enhance its memory management¹ but OCaml cannot use its own
library to do that.  I am not asking WHAT the rules are but a
JUSTIFICATION for them (which you have been incapable of providing so
far).

C.


¹ Apparently the "Please don't implement your own custom memory pool
or free list." does not count if it's implemented by others!

RE: [Caml-list] Is OCaml fast?

[David Allsopp]

Thanassis Tsiodras wrote:
> I apologize beforehand if this is not the forum to ask.
> 
> I am on the fence about whether to learn OCaml or not, and while reading
> an article called "Why OCaml"
> (http://www.cs.ubc.ca/~murphyk/Software/Ocaml/why_ocaml.html), I saw that
> OCaml was praised for the speed of the executables it generates - and was
> referred to, speed-wise, as "second to none", except C and
> C++.

Do remember that speed isn't always everything - there are many applications where it's just not critical. Maintainability, readability, ease-of-use, availability of libraries are large concerns too.

> However, when I actually went to the Language Shootout page suggested in
> the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
> (http://shootout.alioth.debian.org/u32/which-programming-languages-are-
> fastest.php)

As the kinda large number of messages in this thread suggests, benchmarks can be very subjective :o)

The Wiktionary definition of "to benchmark" is "To measure the performance of (an item) relative to another similar item in an impartial scientific manner". IMHO the problem benchmarking "programming languages" (note - "programming languages", not "problems solved in programming languages") lies in the word "similar" in that definition. Reducing an entire programming language's strengths (or weaknesses!) to a single number is just not really realistic - the truth is more complex than one single-precision floating point number (or even an array of them) can describe. (NB. The shootout doesn't claim that the final ranking displayed is anything other than a score of how well the languages did at the various benchmarks given - but a graph like that is easy to interpret erroneously in that way)

> Is it just hype, then? Or am I missing something?

OCaml's big strengths (by no means unique to OCaml, incidentally) for me are its ML legacy - automatic memory (de-)allocation, type inference, static type-checking (i.e. no BizarreRuntimeTypeException's here...) and polymorphism - all things which allow me to write less (and clearer) code to solve a given problem. I also find some of the more exotic features useful - private types and polymorphic variants, for example. I'm forced to program in a few less desirable languages from time to time and sorely miss the basic features of ML. I'd also echo Sylvain's point: I too find the time spent testing and debugging OCaml is much, much lower than, say, equivalent C(++) or Java code. 

Personally, I've found the best way to find out if you like a new programming language is to write a non-trivial program in it (maybe with the occasional assistance of skilled people in its community) and then see what happens when you go back to programming in something you used before.

And if speed really is worrying you, the big thing you may have noticed from the, ahem, slightly animated discussion that's followed is that there are plenty of people on this list who know how to help you identify what's causing the problem and then offer suggestions to tweak it. Personally, in six years of programming in OCaml for a variety of tasks I've never hit the brick wall where I felt the task I was carrying out would have been orders of magnitude faster in another language *and also easy to code in that language*!

HTH,


David

Re: [Caml-list] Re: Is OCaml fast?

[Eray Ozkural]

[-- Attachment #1: Type: text/plain, Size: 1123 bytes --]

On Wed, Nov 24, 2010 at 12:24 PM, Christophe Troestler <
Christophe.Troestler+ocaml@umh.ac.be<Christophe.Troestler%2Bocaml@umh.ac.be>
> wrote:

> On Wed, 24 Nov 2010 06:50:15 +0000, Isaac Gouy wrote:
> >
> > Jeff Meister <nanaki <at> gmail.com> writes:
> >
> > > We know what your rules are for binary-trees; repeating them does
> > > not help.
> >
> > When Christophe TROESTLER wrongly states - "OCaml is not authorized
> > to make use of its very own library!" - he shows that those rules
> > are not known.
>
> Isolating a sentence gives a misleading idea of what I said.  Must I
> really repeat that I was asking WHY is C allowed to use an external
> library to enhance its memory management¹ but OCaml cannot use its own
> library to do that.  I am not asking WHAT the rules are but a
> JUSTIFICATION for them (which you have been incapable of providing so
> far).
>
>


I didn't get it, either.

Best,

-- 
Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo.com/group/ai-philosophy
http://myspace.com/arizanesil http://myspace.com/malfunct

[-- Attachment #2: Type: text/html, Size: 1701 bytes --]

Re: [Caml-list] Is OCaml fast?

[Cedric Cellier]

I was in a similar position than yours two years ago, looking for a
higher level language than C to gain some expressive power but without
willing to sacrifice C's speed. I consulted a lot of benchmarks from
the shootout, and also many webpages similar to the one you refer to,
some praising a language some criticizing it.

The various contenders of the shootout seamed to fall in three categories :
languages that are fast compared to C, but does not reduce much the size
of the source code (C++, java, APS...), languages that allow concise
manipulations of non-trivial data structures (Lisps, MLs, Haskells), and
non-compiled languages that I never considered adopting.

I eliminated Haskell because GHC was only usable on x86 architecture at
that time, so I tried various Lisp at first. Then a friend pointed me to
Ocaml since, according to him, it's easier to cross the gap between C
and Ocaml than between C and anything else. I thus finally tried Ocaml
and, despite my initial apprehensions, I felt myself comfortable with
its syntax after am couple of days.

I am now using it whenever I have the choice to, after many years of C,
mixing with C or ASM when necessary (which is quite easy), because I
feel it's pleasant to use, and that I have still many things to learn
from it, without the bad feeling to waste CPU cycles along the way.

I like to think that if yon consider only speed and terseness of the
code, Ocaml is still amongst the top languages ; probably not the #1 as
stated by the (quite old) webpage you pointed to, but still excellent.

Re: [Caml-list] Is OCaml fast?

[Vincent Aravantinos]

Hi Thanassis,

Le 22 nov. 10 à 14:21, Thanassis Tsiodras a écrit :

> I apologize beforehand if this is not the forum to ask.
>
> (...)
>
> Is it just hype, then? Or am I missing something?

may we know, after all this intense discussion, what is your feeling?

Did this debate enlightened your views?
Did it make arise new questions?
Are you going to learn Ocaml?

At least you know now that Ocaml has a lively community :)
I just hope it did not disgust you...

Cheers,
V. Aravantinos

Re: [Caml-list] Is OCaml fast?

[Thanassis Tsiodras]

On Wed, Nov 24, 2010 at 4:34 PM, Vincent Aravantinos
<vincent.aravantinos@gmail.com> wrote:
> may we know, after all this intense discussion, what is your feeling?

Well... (ducks, wears helmet).

Dr Jon Harrop communicated with me directly (two days ago)... and when
I expressed my lack of faith after reading his "Rise and fall of
OCaml" article at
http://flyingfrogblog.blogspot.com/2010/08/rise-and-fall-of-ocaml.html
, he in fact counter-suggested that F# is now the horse to bet on -
so... what do you guys think?

Over the last couple of days, I've played a lot with ocaml (to be
exact, Linux/ocamlopt, since my interest in the speed of what I make
remains dominant) as well as F# (with Visual Studio 2008). To my
limited understanding, the differences between OCaml and F# are small
- and the benefits of direct access to the .NET ecosystem of libraries
seems to counter the ... uncertain status of OCaml libraries. I just
begun looking into all this, so I could be very wrong, of course - but
I am spoiled rotten with Python's libraries, so not having "batteries
included" in OCaml seemed like quite a problem... until I realized F#
completely covers this.

So, to conclude - what do you guys think about F# ?


>
> Did this debate enlightened your views?
> Did it make arise new questions?
> Are you going to learn Ocaml?
>
> At least you know now that Ocaml has a lively community :)
> I just hope it did not disgust you...
>
> Cheers,
> V. Aravantinos



-- 
What I gave, I have; what I spent, I had; what I kept, I lost. (Old Epitaph)

Re: [Caml-list] Is OCaml fast?

[Oliver Bandel]

Hi,


Zitat von "Thanassis Tsiodras" <ttsiodras@gmail.com>:

[...]
> Over the last couple of days, I've played a lot with ocaml (to be
> exact, Linux/ocamlopt, since my interest in the speed of what I make
> remains dominant)
[...]

OK, then I asume, you will be one of the programmers who
pick out the right datastructures for his applications.

When looking at C because of speed, this is completely ok.

But I would estimate that 80% or 90% of the programmers
just don#t think enough on that topic.

Many people who work as programmers, just pick a language
and a library.... just that library that is the most mainstream
library and start working.

Also many C-programmers I have seen have just done
seemingly simple hacking and not thought about data structures enough.

For simple programs and little amount of input data that's often
ok, but in the history of a program will turn out as bad decision.

So: what are all these benchmarks on different data structures worth,
if C-programmers always stick to arrays, or, maybe, to linked lists?

When I see how Gnome (and I would guess other desktop environments  
will handle it thje same) handles areas of the screen.... that it does  
not use trees that are appropriate for handling two-dimensional  
spatial data, and then experience slow behaviour, then all is said....


I mean: OK, a low performant language will definitley make things  
slower than a higher performant one - when usd with the same algorithms.

But when one does not use intelligent programming, this all does not  
help much.
Then you can use C and be slower than, say, Python (which btw also  
offers some nice datastructures, like Sets and so on).


But maybe what I write here is nothing new to you.

I just wanted to have it mentioned, that benchmarks alone does not  
guarantee fast running programs.

A langauge that supports you to easily implement the datastructures you
would like to use, is much better supporting you then in this respect,
because if it needs toomuch effort to implement such a datastructure,  
lazyness of the programmer might be stronger than the will to use  
beter data structures.... especially if you have to fight with  
dangling pointers on the one hand, where you on the other hand will be  
warned in advance, that you just didn't got the types right and the  
program will not compile then... which yields in fixing errors at  
compilation stage.


Ciao,
   Oliver

Re: [Caml-list] Is OCaml fast?

[Vincent Aravantinos]

Le 24 nov. 10 à 16:30, Thanassis Tsiodras a écrit :

> On Wed, Nov 24, 2010 at 4:34 PM, Vincent Aravantinos
> <vincent.aravantinos@gmail.com> wrote:
>> may we know, after all this intense discussion, what is your feeling?
>
> Well... (ducks, wears helmet).
>
> Dr Jon Harrop communicated with me directly (two days ago)... and when
> I expressed my lack of faith after reading his "Rise and fall of
> OCaml" article at
> http://flyingfrogblog.blogspot.com/2010/08/rise-and-fall-of-ocaml.html
> , he in fact counter-suggested that F# is now the horse to bet on -
> so... what do you guys think?
>
> Over the last couple of days, I've played a lot with ocaml (to be
> exact, Linux/ocamlopt, since my interest in the speed of what I make
> remains dominant) as well as F# (with Visual Studio 2008). To my
> limited understanding, the differences between OCaml and F# are small
> - and the benefits of direct access to the .NET ecosystem of libraries
> seems to counter the ... uncertain status of OCaml libraries. I just
> begun looking into all this, so I could be very wrong, of course - but
> I am spoiled rotten with Python's libraries, so not having "batteries
> included" in OCaml seemed like quite a problem... until I realized F#
> completely covers this.
>
> So, to conclude - what do you guys think about F# ?

I personally do not know much of F#. Particularly because I do not  
have a Windows machine :( We all know here that Jon is very fond of F#  
after having been fond of Ocaml for a while.

The socalled "Ocaml mass exodus" mentionned in Jon's article seems to  
me as as objective as the language shootout benchmarks ;) In  
particular the plot I made myself of the posts to the mailing list is  
definitely not as clear as the one presented in the article (you can  
do it yourself from the figures at <http://caml.inria.fr/pub/ml-archives/caml-list/index.fr.html 
 >). Jon also mentions (in the comments) the figures from Google  
Trends about Ocaml vs F# (<http://www.google.co.uk/trends?q=f 
%23%2Cocaml>). But funnily enough, you can see from the excerpts  
selected by Google that F# deals as much with the language than  
with... ahem, other stuff. Furthermore the plot was already increasing  
before the release of F#, so is this increase really significative? My  
whole point here is not that Jon is wrong or right: maybe he is right,  
I just say that the supposed exodus does not seem significative to me.  
Anyhow I do not "feel" it. I would even say that I feel the contrary  
(and I am not the only one: see Paolo's comment in Jon's article).

Then Jon suggests in his article that that this is due to the  
"inability of Ocaml's GC to [deal with parallelism]". Again, all of us  
here know that Jon has been quite frustratred with this for a while  
(and not only him, cf some very recent thread <http://caml.inria.fr/pub/ml-archives/caml-list/2010/11/642bb03b01227d825e71b0a46ae7e73f.fr.html 
 >). However there are also plenty of guys who are not bothered by  
this (see the very same thread). Choose your camp. On one hand, if you  
go to F# you won't have to choose. On the other hand, this does not  
make F# necessarily faster than Ocaml (as Jon himself proved it: <http://flyingfrogblog.blogspot.com/2010/08/pure-f-now-only-2-slower-than-ocaml.html 
 >).

Of course if you listen to Jon you will be convincend that you should  
choose F#. Many people here will probably react by saying that you  
should choose Ocaml. I am personnally neutral here: in the end, I  
would say that both languages are great (about F# I should say "look  
great" since I never tried it). So whatever is your final choice I  
guess you won't be disappointed.

Cheers,
Vincent

PS: BTW about the complain in Jon's article about the lack of a native  
REPL in ocaml, I think this recent post also answers it very partly: <http://caml.inria.fr/pub/ml-archives/caml-list/2010/11/aef35ed6ad039ce80c5c66175e80fcc5.fr.html 
 >. Probably nothing compared, though, to the F# REPL (trusting Jon on  
that).

Re: Is OCaml fast?

[Isaac Gouy]

Christophe Troestler <Christophe.Troestler+ocaml <at> umh.ac.be> writes:

 
> Isolating a sentence gives a misleading idea of what I said. 

gmane demands "Prune quoted stuff."


> Must I
> really repeat that I was asking WHY is C allowed to use an external
> library to enhance its memory management¹ but OCaml cannot use its own
> library to do that.

Do the OCaml binary-trees programs that are shown not use OCaml memory 

management?

Do the C programs that are shown tune the external library memory management

they use?


> I am not asking WHAT the rules are but a JUSTIFICATION for them (which you
> have been incapable of providing so far).

I feel no need to provide a JUSTIFICATION to you for anything.


> ¹ Apparently the "Please don't implement your own custom memory pool
> or free list." does not count if it's implemented by others!

Many C and C++ programmmers rolled their own custom free list implementations as

part of their program. 

Library memory pool and free list implementations have to be more generally

useful (and probably less efficient) than custom solutions created specifically

for this tiny task.

"Please don't implement your own custom memory pool or free list." means exactly
what it says.

Re: [Caml-list] Re: Is OCaml fast?

[Christophe Raffalli]

[-- Attachment #1.1: Type: text/plain, Size: 3043 bytes --]

Hello,

Here is a test of gctweak.ml on the "now famous" binary-tree shootout bench ...
As you can see it is a 30% speed up which is not too bad, just adding
a file on the compilation command line !

I reattached the file, because I correct a few comments in it ... and a syntax error
that is only visible when not using camlp4 in ocaml-3.11.2 ( { get () with ... } is
valid with camlp4 and invalid without ???) ... Anyway, more work on Gctweak is still needed ...

raffalli@d45-lama:~/Caml$ ocamlopt -o binary_tree gctweak.ml binary_tree.ml ; time ./binary_tree 20
stretch tree of depth 21	 check: -1
2097152	 trees of depth 4	 check: -2097152
524288	 trees of depth 6	 check: -524288
131072	 trees of depth 8	 check: -131072
32768	 trees of depth 10	 check: -32768
8192	 trees of depth 12	 check: -8192
2048	 trees of depth 14	 check: -2048
512	 trees of depth 16	 check: -512
128	 trees of depth 18	 check: -128
32	 trees of depth 20	 check: -32
long lived tree of depth 20	 check: -1

real	0m19.212s
user	0m18.960s
sys	0m0.180s
raffalli@d45-lama:~/Caml$ ocamlopt -o binary_tree binary_tree.ml ; time ./binary_tree 20
stretch tree of depth 21	 check: -1
2097152	 trees of depth 4	 check: -2097152
524288	 trees of depth 6	 check: -524288
131072	 trees of depth 8	 check: -131072
32768	 trees of depth 10	 check: -32768
8192	 trees of depth 12	 check: -8192
2048	 trees of depth 14	 check: -2048
512	 trees of depth 16	 check: -512
128	 trees of depth 18	 check: -128
32	 trees of depth 20	 check: -32
long lived tree of depth 20	 check: -1

real	0m27.484s
user	0m27.270s
sys	0m0.110s

Here is the run with debug := 1 and you see that minor heap size is guessed at
524288, with almost no promoted word (model = 1 means no promoted word)

raffalli@d45-lama:~/Caml$ ocamlopt -o binary_tree gctweak.ml binary_tree.ml ; time ./binary_tree 20
MHS DOUBLED <- 65536 (model 3.996155)
MHS DOUBLED <- 131072 (model 3.000397)
stretch tree of depth 21	 check: -1
MHS DOUBLED <- 262144 (model 2.495375)
MHS DOUBLED <- 524288 (model 1.027698)
2097152	 trees of depth 4	 check: -2097152
524288	 trees of depth 6	 check: -524288
131072	 trees of depth 8	 check: -131072
32768	 trees of depth 10	 check: -32768
8192	 trees of depth 12	 check: -8192
2048	 trees of depth 14	 check: -2048
512	 trees of depth 16	 check: -512
128	 trees of depth 18	 check: -128
32	 trees of depth 20	 check: -32
long lived tree of depth 20	 check: -1

real	0m19.342s
user	0m19.100s
sys	0m0.170s

-- 
Christophe Raffalli
Universite de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tel: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
---------------------------------------------
IMPORTANT: this mail is signed using PGP/MIME
At least Enigmail/Mozilla, mutt or evolution
can check this signature. The public key is
stored on www.keyserver.net
---------------------------------------------

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1.2: gctweak.ml --]
[-- Type: text/x-ocaml; name="gctweak.ml", Size: 6155 bytes --]

open Gc

(* adjustable parameters, should be a functor ? *)
let space_overhead = 100
let gamma = 3.0 (* time in major slice attached to a minor GC  / time for minor GC 
                : should use a real estimation, here just a guess !!! *)
let reactivity = 0.6 (* between 0.5 and 1.0, less or equal than 0.5 is not very reasonable:
                       it is likely to double the minor_heap_size at each major GC, Decrease 
                        if reactivity is not important to you *)
let retraction_coef = 0.9 (* between 0.0 and 1.0. the smaller, the less oscillation
                             in the minor heap size. 0.0: never decrease minor heap *)
let debug = ref 1 (* between 0 and 4, 2 and above is for debugging only *)
let max_minor_heap_size = 1 lsl 25 (* the names is clear, 
                                      rounded to the power of 2 below *) 

let major_heap_increment_ratio = 0.5 (* proportional heap increment ratio *)

(* End of tuning constants *)

(* Justification:
We use a model saying that the time in each GC slice is 

T = K * (m + gamma * r * m * f) where

m = minor heap size (goes away in O())
gamma = define above
r = ratio of promoted word at each minor cycle
f = (space_overhead + 100) / space_overhead used as an estimation
    of free space in major heap after collection
K a time constant

in the sum abobe: 
- K * m is the time in the minor GC
- K * gamma * r * m * f is the time in the major GC slice for each minor GC

If gamma * f is more than 1 (which is likely), it is easy to see that 
increasing m, if it decreases r enough, will both increase overall speed and
time is a GC slice, increasing therefore both speed and reactivity.

More precisely, the model says that this is OK to double the size of the 
minor heap when 1 + gamma * f * r > 2*(1 + gamma * f * r') where r is the ratio 
associated to m and r' is the ratio associated to 2*m.

The code below keeps and update a table of ponderated average value of 
1 + gamma * f * r for all used heap size and tries to make sensible decision
looking at it.
*)

let param = get ()
let _ = set { param with space_overhead = space_overhead }

let main_coef = (* gamma * f *)
  gamma *.
  (float) (space_overhead + 100) /.(float) (space_overhead)

let ratio_double = 1.0 /. reactivity 
let ratio_half = ratio_double /. retraction_coef

(* tranlated log2 of the minor heap size, used at initialization only *)
let index m = 
  let rec fn i m = 
    if m <= 32768 then i else fn (i + 1) (m / 2)
  in
  fn 0 m

(* a table to store 1 + gamma * f * r for each heap size *)
let max_index = index max_minor_heap_size
let model_table = Array.create (max_index+1) None

let old_heap_words = ref (quick_stat ()).heap_words
let old_promoted_words = ref 0.0
let old_minor_collections = ref 0
let minor_heap_size = ref param.minor_heap_size 
let minor_heap_index_size = ref (index param.minor_heap_size)

let _ = create_alarm (fun () -> 
  let s = quick_stat () in

  (* tweak minor heap size *)
  let promoted_words = s.promoted_words in
  let minor_collections = s.minor_collections in
  let delta_promoted_words = promoted_words -. !old_promoted_words in
  let delta_minor_collections = minor_collections - !old_minor_collections in
  old_promoted_words := promoted_words;
  old_minor_collections := minor_collections;
  let ratio = delta_promoted_words /. (float) delta_minor_collections 
                                   /. (float) !minor_heap_size 
  in
  let new_model = 1.0 +. gamma *. ratio in
  let i = !minor_heap_index_size in
  let mean_model = match model_table.(i) with None -> new_model |
    Some r -> (r +. new_model) /. 2.0
  in
  model_table.(i) <- Some mean_model;
  if !debug > 2 then begin
    let i = ref 0 in
    while !i <= max_index && model_table.(!i) <> None do
      match model_table.(!i) with
      | None -> assert false
      | Some r ->  Printf.fprintf stderr "model(%d) = %f - " !i r; incr i
    done;
    Printf.fprintf stderr "\n"; flush stderr;
  end;

  let lower_double, lower_half =
    if i <= 0 then true, false else 
      match model_table.(i-1) with
      | None -> false, true
      | Some r -> let x = 2.0 *. mean_model /. r in
		  (i < max_index) && x < ratio_double, x > ratio_half
  in
  let upper_double, upper_half =
    if i >= max_index then false, true else 
      match model_table.(i+1) with 
      | None -> true, false
      | Some r -> let x = 2.0 *. r /. mean_model in 
		  x < ratio_double, (i > 0) && x > ratio_half
  in
  if !debug > 2 then begin
   Printf.fprintf stderr "ld = %b, lh = %b, ud = %b, uh = %b\n" 
     lower_double lower_half upper_double upper_half;
     flush stderr;
  end;
  if (lower_half && not upper_double) || (upper_half && not lower_double) then begin
    minor_heap_size := !minor_heap_size / 2;
    minor_heap_index_size := i - 1;
    if !debug > 0 then begin
      Printf.fprintf stderr "MHS HALFED <- %d (model %f)\n" !minor_heap_size mean_model;
      flush stderr;
    end;
    set { (get ()) with minor_heap_size = !minor_heap_size }
  end else
  if (lower_double && not upper_half) || (upper_double && not lower_half)  then begin
    minor_heap_size := !minor_heap_size * 2;
    minor_heap_index_size := i + 1;
    if !debug > 0 then begin
      Printf.fprintf stderr "MHS DOUBLED <- %d (model %f)\n" !minor_heap_size mean_model;
      flush stderr; 
    end;
    set { (get ()) with minor_heap_size = !minor_heap_size }
  end else
    if !debug > 1 then begin
      Printf.fprintf stderr "MHS UNCHANGED (model %f) mean_model\n" mean_model;
      flush stderr; 
    end;

  (* tweak major heap increment to be a fraction of major heap size *)
  if !old_heap_words <> s.heap_words then begin
    old_heap_words := s.heap_words; 
    let major_heap_increment = max (124*1024) (int_of_float (float s.heap_words *. major_heap_increment_ratio) )in
(*  Printf.fprintf stderr "MHI <- %d \n" major_heap_increment;
    flush stderr; *)
    set { (get ()) with major_heap_increment = major_heap_increment; }
  end;
)



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: Is OCaml fast?

[Isaac Gouy]

Thanassis Tsiodras <ttsiodras <at> gmail.com> writes:
-snip-
> However, when I actually went to the Language Shootout page suggested
> in the article, I found out that OCaml is not 2nd, it is 13th, behind
> languages like Haskell and C#...
>
(http://shootout.alioth.debian.org/u32/which-programming-languages-are-fastest.php)


You've noted the rank of OCaml (2nd and then 13th)

Can the rank position of OCaml change without the performance of the fastest
programs or the OCaml programs changing?
 
Can rank position in some list indicate how much faster or slower OCaml may be
than the fastest language implementation?

Is the rank position of OCaml stated on that old Doug Bagley scorecard?

http://web.archive.org/web/20010602051335/www.bagley.org/~doug/shootout/craps.shtml

Is the rank position of OCaml stated on the current benchmarks game page you
linked-to?

Re: Is OCaml fast?

[Isaac Gouy]

David Allsopp <dra-news <at> metastack.com> writes:

-snip-
> Reducing an entire programming language's strengths (or
> weaknesses!) to a single number is just not really realistic - the truth is
more complex than one
> single-precision floating point number (or even an array of them) can
describe. (NB. The shootout
> doesn't claim that the final ranking displayed is anything other than a score
of how well the languages did
> at the various benchmarks given - but a graph like that is easy to interpret
erroneously in that way)
-snip-


That summary page 

> > (http://shootout.alioth.debian.org/u32/which-programming-languages-are-
> > fastest.php)

shows box plots (quartiles, outliers, median) of the normalized timing
measurements for each of the tasks, for each of the language implementations.

A graph like that shows some of those language implementations are very fast for
some benchmark programs and very slow for others.

To characterize that as "reducing an entire programming language's strengths (or
weaknesses!) to a single number" seems kind-of misleading.

Especially when the question that page answers is stated 3 times - "Which
programming language implementations have the fastest benchmark programs?"

Re: [Caml-list] Re: Is OCaml fast?

[Ed Keith]

--- On Wed, 11/24/10, Isaac Gouy <igouy2@yahoo.com> wrote:
> 
> 
> > I am not asking WHAT the rules are but a JUSTIFICATION
> for them (which you
> > have been incapable of providing so far).
> 
> I feel no need to provide a JUSTIFICATION to you for
> anything.
> 

Am I to interpret this to mean that the rules are purely arbitrary and capricious with no though behind them? 

I hope I am misinterpreting this statement, but can think of no other interpretation

      -EdK

Ed Keith
e_d_k@yahoo.com

Blog: edkeith.blogspot.com





      

Re: Is OCaml fast?

[Isaac Gouy]

Ed Keith <e_d_k <at> yahoo.com> writes:

> > > I am not asking WHAT the rules are but a JUSTIFICATION
> > for them (which you
> > > have been incapable of providing so far).
> > 
> > I feel no need to provide a JUSTIFICATION to you for
> > anything.
> > 
> 
> Am I to interpret this to mean that the rules are purely arbitrary and
capricious with no though behind them? 

Tendentious.

A great deal of thought has been given to all aspects of the benchmarks game.

> I hope I am misinterpreting this statement, but can think of no other
interpretation

Poverty of imagination.

Re: [Caml-list] Re: Is OCaml fast?

[David Rajchenbach-Teller]

Hey, guys.
Time to stop this, please.

Thanks,
 David

On Nov 24, 2010, at 8:13 PM, Isaac Gouy wrote:

> Ed Keith <e_d_k <at> yahoo.com> writes:
> 
>>>> I am not asking WHAT the rules are but a JUSTIFICATION
>>> for them (which you
>>>> have been incapable of providing so far).
>>> 
>>> I feel no need to provide a JUSTIFICATION to you for
>>> anything.
>>> 
>> 
>> Am I to interpret this to mean that the rules are purely arbitrary and
> capricious with no though behind them? 
> 
> Tendentious.
> 
> A great deal of thought has been given to all aspects of the benchmarks game.
> 
>> I hope I am misinterpreting this statement, but can think of no other
> interpretation
> 
> Poverty of imagination.
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

RE: [Caml-list] Re: Is OCaml fast?

[David Allsopp]

Isaac Gouy wrote:
> David Allsopp <dra-news <at> metastack.com> writes:
> 
> -snip-
> > Reducing an entire programming language's strengths (or
> > weaknesses!) to a single number is just not really realistic - the
> > truth is more complex than one single-precision floating point 
> > number (or even an array of them) can describe. (NB. The shootout
> > doesn't claim that the final ranking displayed is anything other than
> > a score of how well the languages did
> > at the various benchmarks given - but a graph like that is easy to
> > interpret erroneously in that way)
> -snip-
> 
> That summary page
> 
> > > (http://shootout.alioth.debian.org/u32/which-programming-languages-a
> > > re-
> > > fastest.php)
> 
> shows box plots (quartiles, outliers, median) of the normalized timing
> measurements for each of the tasks, for each of the language
> implementations.
> 
> A graph like that shows some of those language implementations are very
> fast for some benchmark programs and very slow for others.

I'm not sure I disputed that anywhere, in fact I think if anything implicitly agreed with it...

> To characterize that as "reducing an entire programming language's
> strengths (or weaknesses!) to a single number" seems kind-of misleading.

I'm not clear how you extracted that from what I said which was that reducing a programming language to a single number (or three numbers, if you want bars) is not a good summary of the *total* strengths of that programming language. I then went on to stress that the combined results from the shootout don't claim to do that, they simply show an interpretation of the combined results from the different benchmarks.

My attempt at a point was that ranking programming languages (that's just "programming languages" not "execution speed of programming languages") is a largely futile activity, because it's just too subjective and hard to quantify in a rigorous manner, and importantly that the shootout doesn't try to do that. The relevance and value of that point to Thanassis' original question is for him to decide...

> Especially when the question that page answers is stated 3 times - "Which
> programming language implementations have the fastest benchmark programs?"

Which part of my statement "The shootout doesn't claim that the final ranking displayed is anything other than a score of how well the languages did at the various benchmarks given" (prefixed with the Latin abbreviation for "Note well") caused you to need to write this?


David

Re: [Caml-list] Re: Is OCaml fast?

[Oliver Bandel]

halting problem discovered practically ;)




Zitat von "David Rajchenbach-Teller" <David.Teller@univ-orleans.fr>:

> Hey, guys.
> Time to stop this, please.
>
> Thanks,
>  David
>
> On Nov 24, 2010, at 8:13 PM, Isaac Gouy wrote:
>
>> Ed Keith <e_d_k <at> yahoo.com> writes:
>>
>>>>> I am not asking WHAT the rules are but a JUSTIFICATION
>>>> for them (which you
>>>>> have been incapable of providing so far).
>>>>
>>>> I feel no need to provide a JUSTIFICATION to you for
>>>> anything.
>>>>
>>>
>>> Am I to interpret this to mean that the rules are purely arbitrary and
>> capricious with no though behind them?
>>
>> Tendentious.
>>
>> A great deal of thought has been given to all aspects of the  
>> benchmarks game.
>>
>>> I hope I am misinterpreting this statement, but can think of no other
>> interpretation
>>
>> Poverty of imagination.
>>
>> _______________________________________________
>> Caml-list mailing list. Subscription management:
>> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
>> Archives: http://caml.inria.fr
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>

Re: [Caml-list] Re: Is OCaml fast?

[Goswin von Brederlow]

Gerd Stolpmann <info@gerd-stolpmann.de> writes:

> Am Dienstag, den 23.11.2010, 17:53 +0000 schrieb Isaac Gouy:
>> Christophe TROESTLER <Christophe.Troestler+ocaml <at> umh.ac.be> writes:
>> 
>> > 
>> > On Tue, 23 Nov 2010 02:03:48 +0000, Isaac Gouy wrote:
>> > >
>> > > > C version : 12.11 secs
>> > > > OCaml version : 47.22 secs
>> > > > OCaml version with GC parameters tuned ("interesting alternative" 
>> > >  section) : 12.67 secs
>> > >
>> > > And of course you know because that GC tuned OCaml program is shown 
>> > >  on the
>> > > benchmarks game website 
>> > > http://shootout.alioth.debian.org/u32/program.php?test=binarytrees& 
>> > > lang=ocaml&id=1
>> > 
>> > Since you are here, please explain why C can use memory pools and vec 
>> > tor instructions but tuning the GC of OCaml ― although it is part of  
>> > the standard library ― is considered an “alternative”.
>> 
>> 
>> You seem to be suggesting that "tuning the GC" is considered "alternative" only
>> for OCaml programs.
>> 
>> You seem to be suggesting that "tuning the GC" is considered "alternative" for
>> every task.
>> 
>> Neither is true.
>> 
>> You seem to be suggesting some kind of equivalence between vector instructions
>> and "tuning the GC".
>> You haven't said why they should be considered equivalent.
>> 
>> Nor have you said why you think C should not be allowed to use memory pools.
>
> Quite easy: because you are comparing results that cannot be compared.
> The reader of this benchmark (binary trees) might have the impression
> that C is generally that fast - however, this would be no longer true if
> these binary trees were used as library in a bigger program where using
> memory pools is inappropriate, e.g. because the data managed by the
> binary trees has an unpredictable lifetime.
>
> I do not say that it is complete nonsense to do this comparison, but
> only that it is more specific than a reader would assume. The innocent
> reader expects a comparison of binary tree performance, not of methods
> of managing memory (and this is it finally). The true name of this test
> should be "manage_many_small_memory_cells_where_pools_are_allowed". (It
> would be actually interesting to compare various versions of this test
> with different memory management methods.)
>
> Gerd

So write an ocaml programm that uses and array or bigarray to pool its
memory. That is the same as C does.

MfG
        Goswin

Re: Is OCaml fast?

[Isaac Gouy]

David Allsopp <dra-news <at> metastack.com> writes:

> > > (NB. The shootout
> > > doesn't claim that the final ranking displayed is anything other than
> > > a score of how well the languages did
> > > at the various benchmarks given - but a graph like that is easy to
> > > interpret erroneously in that way)
> > -snip-


> Which part of my statement "The shootout doesn't claim that the final ranking
displayed is anything other
> than a score of how well the languages did at the various benchmarks given"
(prefixed with the Latin
> abbreviation for "Note well") caused you to need to write this?


The part you haven't included in your quotation ;-)

RE: [Caml-list] Is OCaml fast?

[Jon Harrop]

Boxing and generics with type erasure making it difficult or impossible to
reduce the density of references in the heap. Structs and reified generics‎
can give huge performance improvements. Hash tables and complex arithmetic
are some examples where I often see F# running 5-20× faster than idiomatic
OCaml/Java code.

Cheers,
Jon.

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Erik de Castro Lopo
> Sent: 24 November 2010 01:24
> To: caml-list@inria.fr
> Subject: Re: [Caml-list] Is OCaml fast?
> 
> Jon Harrop wrote:
> 
> > Many of the things that can make OCaml and Java slow were inherited
> > from Lisp. They are, in effect, designs that still bear the burden
> > of dynamic typing despite being statically typed.
> 
> I'm curious, what in your opinion are those things?
> 
> Erik
> --
> ----------------------------------------------------------------------
> Erik de Castro Lopo
> http://www.mega-nerd.com/
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: Is OCaml fast?

[Stefan Monnier]

> Richard's objection, which you dismissed out of hand, was that your
> no-GC-tuning rule is silly in the light of actual uses of garbage
> collected programming languages on modern processors.  It makes your
> results unrealistic, and an unrealistic benchmark is misleading, or at
> best merely useless.

To the extent that this rule is the same for all languages and that most
languages on the shootout are also garbage collected, I think OCaml's
problem with this benchmark do point at a weakness of the current
GC code.

Of course, the shootout could be improved.  E.g. maybe it could allow
extra submissions that break the rules, along with a description of
which rules were broken and how.  Then there could be a "score according
to the rules", then a "score when all gloves are off", together with
some kind of "measure" of what was needed to go from one to the other.
This way people could maybe get a better feel for the languages's
performance and how (and how much) that performance can be affected.
Doesn't seem like an easy undertaking, tho.


        Stefan

Re: Is OCaml fast?

[Isaac Gouy]

Stefan Monnier <monnier <at> iro.umontreal.ca> writes:

-snip-
> Of course, the shootout could be improved.  

Of course, anyone interested can do their own more extensive comparisons for a
couple of language implementations, for example - 

http://blog.marketcetera.com/2007/03/08/java-and-the-computer-language-shootout/


> E.g. maybe it could allow
> extra submissions that break the rules, along with a description of
> which rules were broken and how.  
-snip-


Page search "alternative" on

http://shootout.alioth.debian.org/u32q/performance.php?test=binarytrees

RE: [Caml-list] Re: Is OCaml fast?

[Jon Harrop]

Stefan wrote:
> I think OCaml's
> problem with this benchmark do point at a weakness of the current
> GC code.

A difference but not necessarily a weakness. OCaml makes performance
transparent and gives you the ability to improve it with a high glass
ceiling. I prefer that to hiding the controls and using ad-hoc rules to
tweak things behind your back, rendering performance unpredictable and
making it hard to optimize your code.

Cheers,
Jon.

Re: [Caml-list] Re: Is OCaml fast?

[Fabrice Le Fessant]

[-- Attachment #1: Type: text/plain, Size: 1317 bytes --]

On 11/25/2010 11:12 PM, Jon Harrop wrote:
> Stefan wrote:
>> I think OCaml's problem with this benchmark do point at a weakness of the current
>> GC code.

What makes you think that ?

I have contributed to some of the solutions that you can find there (and 
some other ones were rejected because caching results is not allowed too 
;-) ), and the GC was _NEVER_ a problem !

  (except for binary-trees, but for absurd reasons)

  The main problem was that other languages have bigger standard 
libraries, whereas OCaml has a very small one (just what is needed to 
compile the compiler, actually). In many problems, you could benefit 
from using a very simple shared-memory library (in mandelbrot, the ocaml 
multicore solution has to copy the image in a socket between processes, 
whereas it could just be in a shared memory segment), and in general, 
many solutions could benefit from specialised data structures that are 
provided in other languages by their standard libraries, and from some 
system calls that are currently not in the Unix library.

Anyway, I decided to stop trying to improve the solutions, and work on 
improving the compiler and its libraries instead. It might benefit to 
ocaml ranking in the shootout, but more importantely, it will benefit to 
everybody in the community also.

--Fabrice

[-- Attachment #2: fabrice_le_fessant.vcf --]
[-- Type: text/x-vcard, Size: 384 bytes --]

begin:vcard
fn:Fabrice LE FESSANT
n:LE FESSANT;Fabrice
org:INRIA Saclay -- Ile-de-France;Projet OCamlPro
adr;quoted-printable:;;Parc Orsay Universit=C3=A9 ;Orsay CEDEX;;91893;France
email;internet:fabrice.le_fessant@inria.fr
title;quoted-printable:Charg=C3=A9 de Recherche
tel;work:+33 1 74 85 42 14
tel;fax:+33 1 74 85 42 49 
url:http://fabrice.lefessant.net/
version:2.1
end:vcard

Re: Is OCaml fast?

[Isaac Gouy]

Fabrice Le Fessant <Fabrice.Le_fessant <at> inria.fr> writes:

-snip-
> Anyway, I decided to stop trying to improve the solutions, and work on 
> improving the compiler and its libraries instead. It might benefit to 
> ocaml ranking in the shootout, but more importantely, it will benefit to 
> everybody in the community also.


¡bravo!

Re: [Caml-list] Re: Is OCaml fast?

[Christophe Raffalli]

[-- Attachment #1: Type: text/plain, Size: 1463 bytes --]

Hello,
> To the extent that this rule is the same for all languages and that most
> languages on the shootout are also garbage collected, I think OCaml's
> problem with this benchmark do point at a weakness of the current
> GC code.
This is untrue ... the bintree example, is just bad in OCaml because the
default
value of the minor heap size is the correct value for reactive programs
where you want fast minor GC slice, because they interrupt the program ...

Maybe the Gc library could provide reasonable default settings for both
cases
accessible via a simple function GC.is_reactive : bool -> unit ... And
the shootout
would allow to call such a function ?

Clearly, all benchmarks which just measure speed and space should use a
larger
minor heap size (except if reactive communication between processes is
important in the
benchmark).

More importantly, there is little hope to discover at runtime that a
program is
a reactive one ... In other words there is no hope for a perfect GC for
all situations.
So it is a good thing that OCaml allows tuning the GC.

Moreover, even when you automatically tune the GC (like changing the
number of generations in some modern GC) you do it based on past
observation of the behavior of your program ... and the behavior of the
program can change just after you changed some GC parameters ... So
begin able to help the GC, in a comprehensive way,
is good ...

Cheers,
Christophe


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

Re: Is OCaml fast?

[Stefan Monnier]

>>> I think OCaml's problem with this benchmark do point at a weakness
>>> of the current GC code.
> What makes you think that ?

I don't really understand the question: it was just stating the obvious.
OCaml's GC (including its default settings) is generally very good, but
like all GCs it has its weaknesses.  This is fundamentally unavoidable.
There should be no shame admitting that this particular case hits one of
those weak spots.


        Stefan

Re: [Caml-list] Is OCaml fast?

[Pierre Etchemaïté]

[-- Attachment #1: Type: text/plain, Size: 689 bytes --]

Le Mon, 22 Nov 2010 20:33:34 +0200, Török Edwin <edwintorok@gmail.com> a écrit :

> Not sure what the max should be for the minor heap increase, but based
> on this benchmark increasing size of minor heap never slows down the
> program. Even when size of minor heap exceeds what fits in the cache.
> I guess there is another microbenchmark that can show the opposite
> though.

Another consequence of increasing minor heap size is the latency
introduced by minor heap collections. That said, when does it become
the worse garbage collector latency? (... when heap compaction is
disabled?)
And of course, for many programs such latency matters a lot less than
thoughput...

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

OCaml GC [was Is OCaml fast?]

[Christophe Raffalli]

[-- Attachment #1.1: Type: text/plain, Size: 2735 bytes --]

Le 27/11/10 22:11, Pierre Etchemaïté a écrit :
> Le Mon, 22 Nov 2010 20:33:34 +0200, Török Edwin <edwintorok@gmail.com> a écrit :
>
>> Not sure what the max should be for the minor heap increase, but based
>> on this benchmark increasing size of minor heap never slows down the
>> program. Even when size of minor heap exceeds what fits in the cache.
>> I guess there is another microbenchmark that can show the opposite
>> though.
> Another consequence of increasing minor heap size is the latency
> introduced by minor heap collections. That said, when does it become
> the worse garbage collector latency? (... when heap compaction is
> disabled?)
> And of course, for many programs such latency matters a lot less than
> thoughput...
In fact, an allocated block is visited at most twice by a minor GC and
if promoted (that if visited twice by the minor GC) at least once by the
major GC (when freed because
major GC is mark and sweep) ... So a heap allocated block is traversed
0, 1, 3 or more time by one of the GC ...

So increasing the minor heap size strictly decreases the number of time
a word is examined by one of the GC ...

I think that having two heaps is mainly (only ?) for one reason:
collecting some memory sooner (and therefore using less memory for dead
objects).
This may speed up the program because saving memory decreases cache
misses, but normaly it slows down the
program. But it is still better than one generation GCed more frequently ...

The latency is not directly related to minor collection ... If there was
only the major heap,
OCaml could still to a bit of major-gc cycle at each collection ... This
is a (good) design choice of OCaml
to have a small minor heap that can be collected in a short time by a
simple and fast "stop and copy"
algorithm.

The main problem of the bin-tree benchmark is when you know that an
object will live long enough to be promoted, you can not ask OCaml to
allocate it in the major heap directly. This is why increasing the minor
heap to have all object collected at the first GC works.

A syntax like Node#(x,d,y) to allocate a variant in the major heap
directly together with similar syntax for
tuples, records, ... could be quite useful in OCaml in general for known
long living objects ... An the is the Ancient modules for object that
will live even longer.

Cheers,
Christophe





>
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs


[-- Attachment #1.2: Type: text/html, Size: 4113 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Benedikt Meurer]

Speaking of the OCaml GC in general, wouldn't it make sense to replace the current generational collector with a collector framework that requires less copying in the common case. For example, dividing the heap into two parts, "large object space" and "small object space", where LOS is managed by mark-sweep/mark-compact (could use the current major heap) and SOS is managed by a recent mark-copy/mark-region collector (Immix [1] comes to mind here). That way objects would no longer need to be copied between generations, and using Immix may also reduce cache misses and improve locality of small/medium sized objects (with proper evacuation heuristics). This should be doable with a moderate amount of work, and should require no incompatible changes, while opening the door for concurrent/parallel garbage collection (this would require incompatible changes then, replacing caml_young_ptr/caml_young_limit with TLS vars, etc.).

Benedikt

[1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.144.3640

Re: [Caml-list] Re: Is OCaml fast?

[oliver]

On Thu, Nov 25, 2010 at 11:50:58PM +0100, Fabrice Le Fessant wrote:
[...]
>  The main problem was that other languages have bigger standard
> libraries, whereas OCaml has a very small one (just what is needed
> to compile the compiler, actually). In many problems, you could
> benefit from using a very simple shared-memory library (in
> mandelbrot, the ocaml multicore solution has to copy the image in a
> socket between processes, whereas it could just be in a shared
> memory segment),


...so you work on a shared-mem module?!


> and in general, many solutions could benefit from
> specialised data structures that are provided in other languages by
> their standard libraries, and from some system calls that are
> currently not in the Unix library.
[...]

During the last some releases a lot more unix syscalls were added
and that's fine of course).

Which calls are you missing there?

Ciao,
   Oliver

Re: [Caml-list] Re: Is OCaml fast?

[oliver]

On Sat, Nov 27, 2010 at 04:58:55PM +0100, Christophe Raffalli wrote:
> Hello,
> > To the extent that this rule is the same for all languages and that most
> > languages on the shootout are also garbage collected, I think OCaml's
> > problem with this benchmark do point at a weakness of the current
> > GC code.
> This is untrue ... the bintree example, is just bad in OCaml because the
> default
> value of the minor heap size is the correct value for reactive programs
> where you want fast minor GC slice, because they interrupt the program ...
[...]

And if your program contains both kinds of functionality?
What possible solution would you recommend?


Ciao,
   Oliver

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Eray Ozkural]

[-- Attachment #1: Type: text/plain, Size: 1467 bytes --]

On Sun, Nov 28, 2010 at 4:29 PM, Benedikt Meurer <
benedikt.meurer@googlemail.com> wrote:

> Speaking of the OCaml GC in general, wouldn't it make sense to replace the
> current generational collector with a collector framework that requires less
> copying in the common case. For example, dividing the heap into two parts,
> "large object space" and "small object space", where LOS is managed by
> mark-sweep/mark-compact (could use the current major heap) and SOS is
> managed by a recent mark-copy/mark-region collector (Immix [1] comes to mind
> here). That way objects would no longer need to be copied between
> generations, and using Immix may also reduce cache misses and improve
> locality of small/medium sized objects (with proper evacuation heuristics).
> This should be doable with a moderate amount of work, and should require no
> incompatible changes, while opening the door for concurrent/parallel garbage
> collection (this would require incompatible changes then, replacing
> caml_young_ptr/caml_young_limit with TLS vars, etc.).
>
> Benedikt
>
> [1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.144.3640
>

I was suggesting dealing with small fixed-size objects differently in
another post. This doesn't sound like a bad combination, nice idea :)

Best,

-- 
Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo.com/group/ai-philosophy
http://myspace.com/arizanesil http://myspace.com/malfunct

[-- Attachment #2: Type: text/html, Size: 2031 bytes --]

RE: [Caml-list] OCaml GC [was Is OCaml fast?]

[Jon Harrop]

[-- Attachment #1: Type: text/plain, Size: 2710 bytes --]

I don't understand why this would help here though. Wouldn't that help when
a long-lived structure was single large block but, in this case, the
long-lived structure is a tree composed of many small heap-allocated blocks
and, therefore, they would undergo the same wasteful "allocate in young only
to be copied to old" pathological behaviour?

 

Surely what you really want the ability to spot when a full minor heap
contains mostly live objects and, therefore, make the whole minor heap part
of the major heap, allocate yourself a new minor heap and clear the
remembered sets. IIRC, G1 does something like this.

 

On a related note, When designing HLVM I thought that a mallocs function
that allocated many small blocks of the same size such that they could be
freed individually would be helpful. Another option might be to preallocate
larger numbers of blocks at the same time under the assumption that a thread
allocating many small surviving blocks would continue to do so, as a kind of
pool hybrid.

 

Cheers,

Jon.

 

From: caml-list-bounces@yquem.inria.fr
[mailto:caml-list-bounces@yquem.inria.fr] On Behalf Of Eray Ozkural
Sent: 28 November 2010 18:57
To: Benedikt Meurer
Cc: caml-list@yquem.inria.fr
Subject: Re: [Caml-list] OCaml GC [was Is OCaml fast?]

 

On Sun, Nov 28, 2010 at 4:29 PM, Benedikt Meurer
<benedikt.meurer@googlemail.com> wrote:

Speaking of the OCaml GC in general, wouldn't it make sense to replace the
current generational collector with a collector framework that requires less
copying in the common case. For example, dividing the heap into two parts,
"large object space" and "small object space", where LOS is managed by
mark-sweep/mark-compact (could use the current major heap) and SOS is
managed by a recent mark-copy/mark-region collector (Immix [1] comes to mind
here). That way objects would no longer need to be copied between
generations, and using Immix may also reduce cache misses and improve
locality of small/medium sized objects (with proper evacuation heuristics).
This should be doable with a moderate amount of work, and should require no
incompatible changes, while opening the door for concurrent/parallel garbage
collection (this would require incompatible changes then, replacing
caml_young_ptr/caml_young_limit with TLS vars, etc.).

Benedikt

[1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.144.3640


I was suggesting dealing with small fixed-size objects differently in
another post. This doesn't sound like a bad combination, nice idea :)
 
Best,


-- 
Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo.com/group/ai-philosophy
http://myspace.com/arizanesil http://myspace.com/malfunct


[-- Attachment #2: Type: text/html, Size: 6452 bytes --]

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Benedikt Meurer]

On Nov 28, 2010, at 20:40 , Jon Harrop wrote:

> I don’t understand why this would help here though. Wouldn’t that help when a long-lived structure was single large block but, in this case, the long-lived structure is a tree composed of many small heap-allocated blocks and, therefore, they would undergo the same wasteful “allocate in young only to be copied to old” pathological behaviour?

There is no "young" and no "old" in this scheme. There are two different heaps, one for large objects, one for small (probably 2-8k max object size, so not really small in the sense of OCaml). The small area is managed by Immix, which avoids copying of long-lived objects if they are allocated together with other long-lived objects (likely), or if not evacuates a set of probably related objects to a single chunk (this is somewhat dependent on the evacuation strategy, which will be differnt for OCaml compared to Java), further improving locality. There are simple heuristics to ensure that an object is not evacuated too often (it is already unlikely with the base algorithm), so there will be a lot less copying. One difficulty remains however: the pause times. It would probably be necessary to adopt Immix to a (semi-)incremental style to get the same pause times as with the current GC.

Benedikt

RE: [Caml-list] OCaml GC [was Is OCaml fast?]

[Jon Harrop]

I see. Yes, that sounds like a great idea. How well does Immix cope with
high allocation rates of short-lived objects? Been a while since I read the
Immix paper...

Cheers,
Jon.

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Benedikt Meurer
> Sent: 28 November 2010 20:00
> To: caml-list@yquem.inria.fr
> Subject: Re: [Caml-list] OCaml GC [was Is OCaml fast?]
> 
> 
> On Nov 28, 2010, at 20:40 , Jon Harrop wrote:
> 
> > I don't understand why this would help here though. Wouldn't that
> help when a long-lived structure was single large block but, in this
> case, the long-lived structure is a tree composed of many small heap-
> allocated blocks and, therefore, they would undergo the same wasteful
> "allocate in young only to be copied to old" pathological behaviour?
> 
> There is no "young" and no "old" in this scheme. There are two
> different heaps, one for large objects, one for small (probably 2-8k
> max object size, so not really small in the sense of OCaml). The small
> area is managed by Immix, which avoids copying of long-lived objects if
> they are allocated together with other long-lived objects (likely), or
> if not evacuates a set of probably related objects to a single chunk
> (this is somewhat dependent on the evacuation strategy, which will be
> differnt for OCaml compared to Java), further improving locality. There
> are simple heuristics to ensure that an object is not evacuated too
> often (it is already unlikely with the base algorithm), so there will
> be a lot less copying. One difficulty remains however: the pause times.
> It would probably be necessary to adopt Immix to a (semi-)incremental
> style to get the same pause times as with the current GC.
> 
> Benedikt
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

Re: [Caml-list] Re: Is OCaml fast?

[Christophe Raffalli]

[-- Attachment #1: Type: text/plain, Size: 1703 bytes --]

Le 28/11/2010 19:17, oliver@first.in-berlin.de a écrit :
> On Sat, Nov 27, 2010 at 04:58:55PM +0100, Christophe Raffalli wrote:
>> Hello,
>>> To the extent that this rule is the same for all languages and that most
>>> languages on the shootout are also garbage collected, I think OCaml's
>>> problem with this benchmark do point at a weakness of the current
>>> GC code.
>> This is untrue ... the bintree example, is just bad in OCaml because the
>> default
>> value of the minor heap size is the correct value for reactive programs
>> where you want fast minor GC slice, because they interrupt the program ...
> [...]
> 
> And if your program contains both kinds of functionality?
> What possible solution would you recommend?

Changing the value of the minor heap size at runtime ...
There is no pb with this ...

> 
> 
> Ciao,
>    Oliver
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs

-- 
Christophe Raffalli
Universite de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tel: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
---------------------------------------------
IMPORTANT: this mail is signed using PGP/MIME
At least Enigmail/Mozilla, mutt or evolution
can check this signature. The public key is
stored on www.keyserver.net
---------------------------------------------


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

RE: [Caml-list] Re: Is OCaml fast?

[Jon Harrop]

I assume he means one thread has one behaviour and another has the other behaviour, in which case there certainly is a problem!

Cheers,
Jon.

> -----Original Message-----
> From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> bounces@yquem.inria.fr] On Behalf Of Christophe Raffalli
> Sent: 29 November 2010 07:33
> To: oliver@first.in-berlin.de
> Cc: Caml List
> Subject: Re: [Caml-list] Re: Is OCaml fast?
> 
> Le 28/11/2010 19:17, oliver@first.in-berlin.de a écrit :
> > On Sat, Nov 27, 2010 at 04:58:55PM +0100, Christophe Raffalli wrote:
> >> Hello,
> >>> To the extent that this rule is the same for all languages and that
> most
> >>> languages on the shootout are also garbage collected, I think
> OCaml's
> >>> problem with this benchmark do point at a weakness of the current
> >>> GC code.
> >> This is untrue ... the bintree example, is just bad in OCaml because
> the
> >> default
> >> value of the minor heap size is the correct value for reactive
> programs
> >> where you want fast minor GC slice, because they interrupt the
> program ...
> > [...]
> >
> > And if your program contains both kinds of functionality?
> > What possible solution would you recommend?
> 
> Changing the value of the minor heap size at runtime ...
> There is no pb with this ...
> 
> >
> >
> > Ciao,
> >    Oliver
> >
> > _______________________________________________
> > Caml-list mailing list. Subscription management:
> > http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> > Archives: http://caml.inria.fr
> > Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> > Bug reports: http://caml.inria.fr/bin/caml-bugs
> 
> --
> Christophe Raffalli
> Universite de Savoie
> Batiment Le Chablais, bureau 21
> 73376 Le Bourget-du-Lac Cedex
> 
> tel: (33) 4 79 75 81 03
> fax: (33) 4 79 75 87 42
> mail: Christophe.Raffalli@univ-savoie.fr
> www: http://www.lama.univ-savoie.fr/~RAFFALLI
> ---------------------------------------------
> IMPORTANT: this mail is signed using PGP/MIME
> At least Enigmail/Mozilla, mutt or evolution
> can check this signature. The public key is
> stored on www.keyserver.net
> ---------------------------------------------

Re: [Caml-list] Re: Is OCaml fast?

[oliver]

yes.

On Mon, Nov 29, 2010 at 11:25:52AM -0000, Jon Harrop wrote:
> I assume he means one thread has one behaviour and another has the other behaviour, in which case there certainly is a problem!
> 
> Cheers,
> Jon.
> 
> > -----Original Message-----
> > From: caml-list-bounces@yquem.inria.fr [mailto:caml-list-
> > bounces@yquem.inria.fr] On Behalf Of Christophe Raffalli
> > Sent: 29 November 2010 07:33
> > To: oliver@first.in-berlin.de
> > Cc: Caml List
> > Subject: Re: [Caml-list] Re: Is OCaml fast?
> > 
> > Le 28/11/2010 19:17, oliver@first.in-berlin.de a écrit :
> > > On Sat, Nov 27, 2010 at 04:58:55PM +0100, Christophe Raffalli wrote:
> > >> Hello,
> > >>> To the extent that this rule is the same for all languages and that
> > most
> > >>> languages on the shootout are also garbage collected, I think
> > OCaml's
> > >>> problem with this benchmark do point at a weakness of the current
> > >>> GC code.
> > >> This is untrue ... the bintree example, is just bad in OCaml because
> > the
> > >> default
> > >> value of the minor heap size is the correct value for reactive
> > programs
> > >> where you want fast minor GC slice, because they interrupt the
> > program ...
> > > [...]
> > >
> > > And if your program contains both kinds of functionality?
> > > What possible solution would you recommend?
> > 
> > Changing the value of the minor heap size at runtime ...
> > There is no pb with this ...
> > 
> > >
> > >
> > > Ciao,
> > >    Oliver
> > >
> > > _______________________________________________
> > > Caml-list mailing list. Subscription management:
> > > http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> > > Archives: http://caml.inria.fr
> > > Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> > > Bug reports: http://caml.inria.fr/bin/caml-bugs
> > 
> > --
> > Christophe Raffalli
> > Universite de Savoie
> > Batiment Le Chablais, bureau 21
> > 73376 Le Bourget-du-Lac Cedex
> > 
> > tel: (33) 4 79 75 81 03
> > fax: (33) 4 79 75 87 42
> > mail: Christophe.Raffalli@univ-savoie.fr
> > www: http://www.lama.univ-savoie.fr/~RAFFALLI
> > ---------------------------------------------
> > IMPORTANT: this mail is signed using PGP/MIME
> > At least Enigmail/Mozilla, mutt or evolution
> > can check this signature. The public key is
> > stored on www.keyserver.net
> > ---------------------------------------------
> 
> 

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Benedikt Meurer]

On Nov 29, 2010, at 00:34 , Jon Harrop wrote:

> I see. Yes, that sounds like a great idea. How well does Immix cope with
> high allocation rates of short-lived objects? Been a while since I read the
> Immix paper...

In theory this should be handled more efficiently compared to a generational collector (given an appropriate evacuation heuristic), since there is less copying and short-lived objects tend to reuse the "hot" memory of objects that died previously (similar to the minor heap now), while survivers benefit from better compaction (without the need for a seperate mark-compact phase as implemented in the current major GC).

The timings for Java are impressive, but the allocation rates in Java tend to be different from those typically found in functional programs. There was also a GSoC project to implement Immix for GHC [1], but they seem to have used Immix in a generational setting, with mixed results (my knowledge of the GHC GC is limited). 

> Cheers,
> Jon.

greets,
Benedikt

[1] http://hackage.haskell.org/trac/ghc/wiki/Commentary/Rts/Storage/GC/Immix

Re: [Caml-list] Re: Is OCaml fast?

[Gerd Stolpmann]

Am Sonntag, den 28.11.2010, 19:14 +0100 schrieb
oliver@first.in-berlin.de:
> On Thu, Nov 25, 2010 at 11:50:58PM +0100, Fabrice Le Fessant wrote:
> [...]
> >  The main problem was that other languages have bigger standard
> > libraries, whereas OCaml has a very small one (just what is needed
> > to compile the compiler, actually). In many problems, you could
> > benefit from using a very simple shared-memory library (in
> > mandelbrot, the ocaml multicore solution has to copy the image in a
> > socket between processes, whereas it could just be in a shared
> > memory segment),
> 
> 
> ...so you work on a shared-mem module?!

Don't know what Fabrice is referring to, but at least I work on a
multicore-enabling library:

https://godirepo.camlcity.org/svn/lib-ocamlnet2/trunk/code/src/netmulticore/

This is work in progress and highly experimental. What's currently
available:

- managing processes and resources like files, shared memory objects 
  etc.
- support for message passing via Netcamlbox (another library)
- low-level only so far: shared memory, including copying Ocaml values
  to and from shm
- low-level only so far: kernel semaphores
- a Netmulticore process is also a Netplex container, so mixing with
  Netplex-managed servers is possible. Also, the Netplex plugins 
  are available (semaphores, mutexes, global variables), but these
  are relatively slow

I've also written a few examples:

https://godirepo.camlcity.org/svn/lib-ocamlnet2/trunk/code/examples/multicore/

Don't expect wonders, though. For instance, the port of the chameneos
game is based on message passing, but this is by design slower than
direct use of shared memory (and this game mostly consists of
synchronization, and there is not any computation involved where
multicore would be an advantage).

The further direction is the addition of more primitives, especially for
managing shared memory. The difficult here is that there is not any
support in the core language, and I have to work around that. This is
based on

val Netsys_mem.init_value 
      ?targetaddr:nativeint -> 
      ?target_custom_ops:(string * custom_ops) list ->
      memory -> int -> 'a -> init_value_flag list -> (int * int)

where type memory = (char,Bigarray.int8_unsigned_elt,Bigarray.c_layout) Bigarray.Array1.t

This means shared memory is handled just as a bigarray of chars that is
mapped into the address spaces of several processes. The init_value
function copies an arbitrary Ocaml value to such a bigarray  - in the
same way as Marshal.to_string - only that it is "to_bigarray", and that
the same representation is used as for normal values (so you can access
the copied values directly). This copying is quite time-consuming: you
have to create the Ocaml value in normal memory first, and then to use a
quite expensive generic copying machinery to get it to shared memory. It
would be more elegant if there was a way to instruct ocamlopt so that
code is emitted that creates the value directly in a user-supplied
memory area.

The arguments targetaddr and target_custom_ops point to another
difficulty: For certain uses of shared memory you cannot ensure that all
processes map the area to the same address. Because of this, there is
support for filling bigarrays so that the addresses are offset for a
different final mapping address. Netcamlbox uses this feature - the
sending process can map the shared bigarray to any address, and
nevertheless format a message with addresses that are right for the
receiving process.

In general, shared memory is difficult to add in an add-on library.
However, some lessons can be learned, and maybe this results in some
"plot" for adding better support in the core language.

Gerd

> 
> 
> > and in general, many solutions could benefit from
> > specialised data structures that are provided in other languages by
> > their standard libraries, and from some system calls that are
> > currently not in the Unix library.
> [...]
> 
> During the last some releases a lot more unix syscalls were added
> and that's fine of course).
> 
> Which calls are you missing there?
> 
> Ciao,
>    Oliver
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
> 


-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

Threading and SharedMem (Re: [Caml-list] Re: Is OCaml fast?)

[Oliver Bandel]

Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:

> Am Sonntag, den 28.11.2010, 19:14 +0100 schrieb
> oliver@first.in-berlin.de:
>> On Thu, Nov 25, 2010 at 11:50:58PM +0100, Fabrice Le Fessant wrote:
>> [...]
>> >  The main problem was that other languages have bigger standard
>> > libraries, whereas OCaml has a very small one (just what is needed
>> > to compile the compiler, actually). In many problems, you could
>> > benefit from using a very simple shared-memory library (in
>> > mandelbrot, the ocaml multicore solution has to copy the image in a
>> > socket between processes, whereas it could just be in a shared
>> > memory segment),
>>
>>
>> ...so you work on a shared-mem module?!
>
> Don't know what Fabrice is referring to, but at least I work on a
> multicore-enabling library:
>
> https://godirepo.camlcity.org/svn/lib-ocamlnet2/trunk/code/src/netmulticore/
>
> This is work in progress and highly experimental. What's currently
> available:
>
> - managing processes and resources like files, shared memory objects
>   etc.
> - support for message passing via Netcamlbox (another library)
> - low-level only so far: shared memory, including copying Ocaml values
>   to and from shm
[...]

You use shared mem(?), but you link only to *.ml files,
and I see no *.c there.

How can this be done?

At least not via the libs that are shipped with OCaml?!

I would have expected some *.c for the shared mem part and
the creation of Caml-values....


Ciao,
    Oliver

P.S.: OCaml also provides a Thread-Lib, which seems to use pthread-lib.
       Normally this should help in making things possible to run on multiple
       cores. What are the restrictions  that this does not run that way?
       Somehow... when all values are handled via one GC, then those threads
       are somehow bound together, but on the other side, it works threaded,
       and consumer-worker pipes and such stuff can be used.
       So... somehow the GC seems to be the point, where the show will be
       stopped? (Anyone who has looked inside OCaml here more detailed?)

Re: Threading and SharedMem (Re: [Caml-list] Re: Is OCaml fast?)

[Gerd Stolpmann]

Am Montag, den 29.11.2010, 17:12 +0100 schrieb Oliver Bandel:
> Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:
> 
> > Am Sonntag, den 28.11.2010, 19:14 +0100 schrieb
> > oliver@first.in-berlin.de:
> >> On Thu, Nov 25, 2010 at 11:50:58PM +0100, Fabrice Le Fessant wrote:
> >> [...]
> >> >  The main problem was that other languages have bigger standard
> >> > libraries, whereas OCaml has a very small one (just what is needed
> >> > to compile the compiler, actually). In many problems, you could
> >> > benefit from using a very simple shared-memory library (in
> >> > mandelbrot, the ocaml multicore solution has to copy the image in a
> >> > socket between processes, whereas it could just be in a shared
> >> > memory segment),
> >>
> >>
> >> ...so you work on a shared-mem module?!
> >
> > Don't know what Fabrice is referring to, but at least I work on a
> > multicore-enabling library:
> >
> > https://godirepo.camlcity.org/svn/lib-ocamlnet2/trunk/code/src/netmulticore/
> >
> > This is work in progress and highly experimental. What's currently
> > available:
> >
> > - managing processes and resources like files, shared memory objects
> >   etc.
> > - support for message passing via Netcamlbox (another library)
> > - low-level only so far: shared memory, including copying Ocaml values
> >   to and from shm
> [...]
> 
> You use shared mem(?), but you link only to *.ml files,
> and I see no *.c there.

cd ../netsys

it's part of a larger package

> 
> How can this be done?
> 
> At least not via the libs that are shipped with OCaml?!
> 
> I would have expected some *.c for the shared mem part and
> the creation of Caml-values....
> 
> 
> Ciao,
>     Oliver
> 
> P.S.: OCaml also provides a Thread-Lib, which seems to use pthread-lib.
>        Normally this should help in making things possible to run on multiple
>        cores. What are the restrictions  that this does not run that way?
>        Somehow... when all values are handled via one GC, then those threads
>        are somehow bound together, but on the other side, it works threaded,
>        and consumer-worker pipes and such stuff can be used.
>        So... somehow the GC seems to be the point, where the show will be
>        stopped? (Anyone who has looked inside OCaml here more detailed?)

Quite easy: there is a global lock, and when Ocaml code runs, this lock
must be acquired. So only one of the pthreads can have this lock, and so
only one pthread can run Ocaml code.

The reason is that memory management is not thread-safe.

Gerd

> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
> 


-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------

Re: Threading and SharedMem (Re: [Caml-list] Re: Is OCaml fast?)

[Oliver Bandel]

Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:

> Am Montag, den 29.11.2010, 17:12 +0100 schrieb Oliver Bandel:
>> Zitat von "Gerd Stolpmann" <info@gerd-stolpmann.de>:
>>
>> > Am Sonntag, den 28.11.2010, 19:14 +0100 schrieb
>> > oliver@first.in-berlin.de:
>> >> On Thu, Nov 25, 2010 at 11:50:58PM +0100, Fabrice Le Fessant wrote:
>> >> [...]
>> >> >  The main problem was that other languages have bigger standard
>> >> > libraries, whereas OCaml has a very small one (just what is needed
>> >> > to compile the compiler, actually). In many problems, you could
>> >> > benefit from using a very simple shared-memory library (in
>> >> > mandelbrot, the ocaml multicore solution has to copy the image in a
>> >> > socket between processes, whereas it could just be in a shared
>> >> > memory segment),
>> >>
>> >>
>> >> ...so you work on a shared-mem module?!
>> >
>> > Don't know what Fabrice is referring to, but at least I work on a
>> > multicore-enabling library:
>> >
>> >  
>> https://godirepo.camlcity.org/svn/lib-ocamlnet2/trunk/code/src/netmulticore/
>> >
>> > This is work in progress and highly experimental. What's currently
>> > available:
>> >
>> > - managing processes and resources like files, shared memory objects
>> >   etc.
>> > - support for message passing via Netcamlbox (another library)
>> > - low-level only so far: shared memory, including copying Ocaml values
>> >   to and from shm
>> [...]
>>
>> You use shared mem(?), but you link only to *.ml files,
>> and I see no *.c there.
>
> cd ../netsys
>
> it's part of a larger package

ah, ok. :)


>
>>
>> How can this be done?
>>
>> At least not via the libs that are shipped with OCaml?!
>>
>> I would have expected some *.c for the shared mem part and
>> the creation of Caml-values....
>>
>>
>> Ciao,
>>     Oliver
>>
>> P.S.: OCaml also provides a Thread-Lib, which seems to use pthread-lib.
>>        Normally this should help in making things possible to run  
>> on multiple
>>        cores. What are the restrictions  that this does not run that way?
>>        Somehow... when all values are handled via one GC, then those threads
>>        are somehow bound together, but on the other side, it works threaded,
>>        and consumer-worker pipes and such stuff can be used.
>>        So... somehow the GC seems to be the point, where the show will be
>>        stopped? (Anyone who has looked inside OCaml here more detailed?)
>
> Quite easy: there is a global lock, and when Ocaml code runs, this lock
> must be acquired. So only one of the pthreads can have this lock,
[...]

Aha, ok.
Thanks for the details.

Wouldn't it be possible for each thread to have it's own GC,
and letting the global lock on global GC only be used, when
global variables are touched?
Then this could be added to the Threads-module...

Ciao,
    Oliver

Re: [Caml-list] Re: Is OCaml fast?

[Christophe Raffalli]

[-- Attachment #1: Type: text/plain, Size: 700 bytes --]

Le 29/11/10 12:44, oliver@first.in-berlin.de a écrit :
> yes.
>
> On Mon, Nov 29, 2010 at 11:25:52AM -0000, Jon Harrop wrote:
>> I assume he means one thread has one behaviour and another has the other behaviour, in which case there certainly is a problem!

Then, I do not know what you can do ... Each thread does not have its
own minor heap, so the size is the same for all ... probably in this
case you want to favor reactivity and keep the minor heap small ?

Anyway, In this situation, even with the default size, with Posix
thread, I do have problems with too slow switching with one thread
computing while the other is in charge of the interface ...

Cheers,
Christophe




[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Török Edwin]

On Sun, 28 Nov 2010 15:29:08 +0100
Benedikt Meurer <benedikt.meurer@googlemail.com> wrote:

> Speaking of the OCaml GC in general, wouldn't it make sense to
> replace the current generational collector with a collector framework
> that requires less copying in the common case. 

Even without changing the GC algorithm, I think it would be useful if we
could minimize the amount of "useless work" major collections do in the
current GC.
I profiled the binary trees with valgrind's callgrind, and it turns out
that most of the time for each minor collection is spent in
caml_major_collection_slice:
http://www.pasteall.org/pic/show.php?id=7194

This seems to be in concordance with the "smaller minor heap => more
minor collections => slower program" observation, since it is:
smaller minor heap => more minor collections => more major slices =>
major slices can't collect long-lived objects => slower program (long
lived objects are sweeped too many times).
So more minor collections => higher cost for a major slice

I think OCaml's GC should take into account how successful the last
major GC was (how many words it freed), and adjust speed accordingly:
if we know that the major slice will collect next to nothing there is
no point in wasting time and running it.

So this formula should probably be changed:
  p = (double) caml_allocated_words * 3.0 * (100 + caml_percent_free)
      / Wsize_bsize (caml_stat_heap_size) / caml_percent_free / 2.0;

Probably to something that also does:
  p =  p * major_slice_successrate

where successrate is how many words the GC succeeded in collecting the
last major slice (or last major collection).

I'm not familiar with OCaml's GC, but it looks like it doesn't keep
track of these stats during a collection, or does it?

P.S.: It would be good if this internal speed percentage was tunable
from the Gc module, at least by a constant factor.

Best regards,
--Edwin

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Damien Doligez]

On 2010-11-29, at 23:27, Török Edwin wrote:

> This seems to be in concordance with the "smaller minor heap => more
> minor collections => slower program" observation, since it is:
> smaller minor heap => more minor collections => more major slices =>
> major slices can't collect long-lived objects => slower program (long
> lived objects are sweeped too many times).
> So more minor collections => higher cost for a major slice

This is a bit naive, because the size of a major slice depends on the
amount of data promoted by the corresponding minor GC.  A smaller minor
heap promotes less data each time, so you get more major slices, but
they are smaller.  The total amount of work done by the major GC doesn't
change because of that.  It only changes because a bigger minor heap
gives more time for data to die before being promoted to the major heap.

> I think OCaml's GC should take into account how successful the last
> major GC was (how many words it freed), and adjust speed accordingly:
> if we know that the major slice will collect next to nothing there is
> no point in wasting time and running it.
> 
> So this formula should probably be changed:
>  p = (double) caml_allocated_words * 3.0 * (100 + caml_percent_free)
>      / Wsize_bsize (caml_stat_heap_size) / caml_percent_free / 2.0;
> 
> Probably to something that also does:
>  p =  p * major_slice_successrate


The success rate is already taken into account by the major GC.  In fact
a target success rate is set by the user: it is
   caml_percent_free / (100 + caml_percent_free)
and the major GC adjusts its speed according to this target.  If the
target is not met, the free list is emptied faster than the GC can
replenish it, so it gets depleted at some point and the major heap
is then extended to satisfy allocation requests.  The bigger major
heap then helps the major GC meet its target, because the success rate
is simply (heap_size - live_data) / heap_size, and that gets higher
as heap_size increases.

I didn't do the math, but I think your modification would make the
major heap size increase without bound.

-- Damien

Re: [Caml-list] OCaml GC [was Is OCaml fast?]

[Török Edwin]

[-- Attachment #1: Type: text/plain, Size: 3951 bytes --]

On Thu, 2 Dec 2010 13:57:23 +0100
Damien Doligez <Damien.Doligez@inria.fr> wrote:

> 
> On 2010-11-29, at 23:27, Török Edwin wrote:
> 
> > This seems to be in concordance with the "smaller minor heap => more
> > minor collections => slower program" observation, since it is:
> > smaller minor heap => more minor collections => more major slices =>
> > major slices can't collect long-lived objects => slower program
> > (long lived objects are sweeped too many times).
> > So more minor collections => higher cost for a major slice
> 
> This is a bit naive, because the size of a major slice depends on the
> amount of data promoted by the corresponding minor GC.  A smaller
> minor heap promotes less data each time, so you get more major
> slices, but they are smaller.  The total amount of work done by the
> major GC doesn't change because of that.  It only changes because a
> bigger minor heap gives more time for data to die before being
> promoted to the major heap.

Thanks for the explanation. 

Is there a way I could use more than 2 bits for the color?
I thought to try excluding some objects from collection if they failed
to get collected for a few cycles (i.e. just mark them black from the
start). Tried modifying Wosize_hd, and Make_header, but it is not
enough, anything else I missed?

> 
> > I think OCaml's GC should take into account how successful the last
> > major GC was (how many words it freed), and adjust speed
> > accordingly: if we know that the major slice will collect next to
> > nothing there is no point in wasting time and running it.
> > 
> > So this formula should probably be changed:
> >  p = (double) caml_allocated_words * 3.0 * (100 + caml_percent_free)
> >      / Wsize_bsize (caml_stat_heap_size) / caml_percent_free / 2.0;
> > 
> > Probably to something that also does:
> >  p =  p * major_slice_successrate
> 
> 
> The success rate is already taken into account by the major GC.  In
> fact a target success rate is set by the user: it is
>    caml_percent_free / (100 + caml_percent_free)

Thanks, that is probably what I had in mind, but adjusting it doesn't
work as I initially expected.

> and the major GC adjusts its speed according to this target.  If the
> target is not met, the free list is emptied faster than the GC can
> replenish it, so it gets depleted at some point and the major heap
> is then extended to satisfy allocation requests.  The bigger major
> heap then helps the major GC meet its target, because the success rate
> is simply (heap_size - live_data) / heap_size, and that gets higher
> as heap_size increases.
> 
> I didn't do the math, but I think your modification would make the
> major heap size increase without bound.

Yeah, I've been experimenting, and unless I put a lower bound on
'p' (like half of initial one) it used way too much memory (sure it was
also faster, but then it is also faster by not running the GC at all
which is not good).

Which parameter do you think could be automatically tuned to accomodate
bursts of allocations? Maybe major_heap_increment (based on recent
history of heap expansions, and promoted words)?

BTW I was also looking at ways to speed up mark_slice and sweep_slice,
(which is not easy since it is mostly optimized already).
Found a place where GCC is not smart enough though. Attached patch
improves GC sweep_slice by 3% - 10% in ocamlopt, depending on CPU (3%
Intel, 10% AMD Phenom II X6). Should I open a bug and attach it?

Using gcc -O3 gives would sometimes give the same benefits, but I
don't really trust -O3, -O2 is as far as I would go. See here for all
the details: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46763
[speaking of which why is the GC compiled with -O, is -O2 too risky
too?]

P.S.: minor_heap_size got bumped in 3.12 branch, so it is now greater
than major_heap_increment. Is that intended?

Best regards,
--Edwin

[-- Attachment #2: gc_opt.patch --]
[-- Type: text/x-patch, Size: 2499 bytes --]

diff -ru /home/edwin/ocaml-3.12.0+rc1/byterun//major_gc.c ./major_gc.c
--- /home/edwin/ocaml-3.12.0+rc1/byterun//major_gc.c	2009-11-04 14:25:47.000000000 +0200
+++ ./major_gc.c	2010-12-02 17:02:38.000000000 +0200
@@ -286,21 +286,25 @@
 {
   char *hp;
   header_t hd;
+  /* speed opt: keep global in local var */
+  char *gc_sweep_hp = caml_gc_sweep_hp;
 
   caml_gc_message (0x40, "Sweeping %ld words\n", work);
   while (work > 0){
-    if (caml_gc_sweep_hp < limit){
-      hp = caml_gc_sweep_hp;
+    if (gc_sweep_hp < limit){
+      hp = gc_sweep_hp;
       hd = Hd_hp (hp);
       work -= Whsize_hd (hd);
-      caml_gc_sweep_hp += Bhsize_hd (hd);
+      gc_sweep_hp += Bhsize_hd (hd);
+      PREFETCH_READ_NT(gc_sweep_hp);
       switch (Color_hd (hd)){
       case Caml_white:
+	caml_gc_sweep_hp = gc_sweep_hp;
         if (Tag_hd (hd) == Custom_tag){
           void (*final_fun)(value) = Custom_ops_val(Val_hp(hp))->finalize;
           if (final_fun != NULL) final_fun(Val_hp(hp));
         }
-        caml_gc_sweep_hp = caml_fl_merge_block (Bp_hp (hp));
+        gc_sweep_hp = caml_fl_merge_block (Bp_hp (hp));
         break;
       case Caml_blue:
         /* Only the blocks of the free-list are blue.  See [freelist.c]. */
@@ -311,7 +315,7 @@
         Hd_hp (hp) = Whitehd_hd (hd);
         break;
       }
-      Assert (caml_gc_sweep_hp <= limit);
+      Assert (gc_sweep_hp <= limit);
     }else{
       chunk = Chunk_next (chunk);
       if (chunk == NULL){
@@ -320,11 +324,12 @@
         work = 0;
         caml_gc_phase = Phase_idle;
       }else{
-        caml_gc_sweep_hp = chunk;
+        gc_sweep_hp = chunk;
         limit = chunk + Chunk_size (chunk);
       }
     }
   }
+  caml_gc_sweep_hp = gc_sweep_hp;
 }
 
diff -ru /home/edwin/ocaml-3.12.0+rc1/byterun//memory.h ./memory.h
--- /home/edwin/ocaml-3.12.0+rc1/byterun//memory.h	2008-12-03 20:09:09.000000000 +0200
+++ ./memory.h	2010-12-02 17:06:12.000000000 +0200
@@ -215,6 +215,16 @@
   #define CAMLunused
 #endif
 
+/* non-temporal prefetch for read (it need not be left in the cache after the access) */
+#if defined (__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 1))
+  #define PREFETCH_READ_NT(addr) __builtin_prefetch((addr), 0, 0)
+  #define PREFETCH_READ(addr) __builtin_prefetch((addr), 0, 3)
+#else
+  #define PREFETCH_READ_NT(addr)
+  #define PREFETCH_READ(addr)
+#endif
+
+
 #define CAMLxparam1(x) \
   struct caml__roots_block caml__roots_##x; \
   CAMLunused int caml__dummy_##x = ( \

Re: [Caml-list] optimizing caml_modify [was OCaml GC [was Is OCaml fast?]]

[ygrek]

On Thu, 2 Dec 2010 17:07:59 +0200
Török Edwin <edwintorok@gmail.com> wrote:

> BTW I was also looking at ways to speed up mark_slice and sweep_slice,
> (which is not easy since it is mostly optimized already).
> Found a place where GCC is not smart enough though. Attached patch
> improves GC sweep_slice by 3% - 10% in ocamlopt, depending on CPU (3%
> Intel, 10% AMD Phenom II X6). Should I open a bug and attach it?

BTW, while we are on this topic, why the following is not in upstream yet?

http://eigenclass.org/R2/writings/optimizing-caml_modify

Looks like a clear win-win without drawbacks (increased code size shouldn't be significant
cause Modify is only used in a couple of places). What do you think?

PS I was hoping Mauricio will push it to bugtracker or start discussion but apparently this didn't happen so
here we go - stealing the flag, sorry :)

-- 
 ygrek
 http://ygrek.org.ua