[9fans] hyperthreading

[9fans] hyperthreading

[andrey mirtchovski]

quick question:

will Plan 9 work with hyperthreading processors? will it recognize them as
dual-cpus?

can you share experience with those?

thanks: andrey

Re: [9fans] hyperthreading

[jmk]

On Fri Dec 26 18:06:45 EST 2003, mirtchov@cpsc.ucalgary.ca wrote:
> quick question:
>
> will Plan 9 work with hyperthreading processors? will it recognize them as
> dual-cpus?
>
> can you share experience with those?
>
> thanks: andrey

Depends. In 2002 we bought a dual-processor Xeon machine with
hyperthreading. After enabling the hyperthreading option in the
BIOS setup Plan 9 detects and uses 4 processors. (I don't think
it makes much, if any, improvement to the system's performance).

Earlier this month we tried a new single-processor P4 system
with hyperthreading. Even with the BIOS option set the BIOS MP
table only contains a single processor entry. I haven't bothered
yet to look to see if there's some BIOS problem or if there's
some other way to take advantage of hyperthreading on this
system.

[9fans] fossil/venti

[vdharani]

Hi,

I am currently reinstalling my plan 9 machine with fossil/venti.

I did a fossil based installation and also setup Venti server. At one
point 'snap -a' also succeeded. When I rebooted the machine, boot process
fails with the message: "boot: /386/init: not connected with venti server".

I know I am clealy missing something but not sure what it is. Here are my
questions:

- can venti and fossil be in the same machine?
- where do I specify venti server details which fossil uses?
- does fossil clear the data blocks once it has saved to venti? is venti
needed while booting the system? if so, who starts it? how does it get
started while booting? do i need to specify in any config file?
- any kernel modification is needed?
- I didnt configure the network adapter yet. Is that a problem for booting?

Any help appreciated.

Thanks
dharani

Re: [9fans] fossil/venti

[okamoto]

> fails with the message: "boot: /386/init: not connected with venti server".

I'm seeing this sometime here, too.
You can getout, and then, re boot fossil agian, and it'll make
no harm so far.

Kenji

Re: [9fans] fossil/venti

[mirtchov]

> I will be happy if someone can give your suggestions. Any help appreciated.

remove 'venti=' from plan9.ini and see if it boots ok.

if the venti resides on the same disk as the fossil (or the same
system that is currently booting) then set the venti to point to a
disk directly, instead of using an IP address.  here's what my machine
uses:

	venti='#S/sdC0/arenas'

as for fossil+venti being more complex than kfs -- venti gives you
more than anything kfs ever did, so it is expected that its
configuration is more complex since it's so site-specific.  if,
instead, you had skipped the venti config and just did a standalone
fossil install you'ld find it just as easy as kfs...

andrey

ps: best thing to do once you get it working is to tear it down and
rebuild it again :)

Re: [9fans] fossil/venti

[vdharani]

hi

>> fails with the message: "boot: /386/init: not connected with venti
>> server".
>
> I'm seeing this sometime here, too.
> You can getout, and then, re boot fossil agian, and it'll make
> no harm so far.

As I read and re-read the postings (including your posting
titled 'building small Plan 9 network at home') in 9fans list, I am able
to guess what is happening. Looks like fossil is not able to connect to
venti probably because venti is not running, network is not up (or not
configured), or fossil doesnt know the location of venti.

After a couple of tries, these are the things I found (please correct me
if I am wrong):

- plan 9 installation is getting complex. the easiest one i did was the
first free version (just a standalone system with kfs). as features are
added, it getting tougher.

- I tried to find out where installation script stores 'flproto' file but
I couldnt. I also couldnt find out where Venti's address is stored that
fossil can use.

- i think kfs file system is needed anyway atleast till fossil kernel is
built and the config files are modified accordingly. it is not possible or
easy to setup fossil/venti without kfs. am i right?

- looks like if the IP address or network changes, fossil may hesitate to
start. So having 'localhost' configuration for fossil/venti communication
may be useful.

- The Wiki document titled 'Setting Up Fossil' is useful but if I am
right, it was written for a fresh fossil installation (instead of
installation-process-assisted fossil installation). So I am not sure what
is done already and what needs to be done. In particular, I am struggling
to get read-write permission on any file (like disk/kfscmd allow).

I will be happy if someone can give your suggestions. Any help appreciated.

I used to feel I am just a end-user to Plan 9. This attempt made me even
more worried. :-(

regards
dharani

Re: [9fans] hyperthreading

[jmk]

I took a quick look at the P4 hyperthreading docs. It looks like we'd
have to use the ACPI tables rather than the BIOS MPS table to correctly
initialise the extra logical CPUs if the MPS table doesn't lie and pretend
the logical CPUs are physical ones. That's more work than anyone here has
time for right now. Some ACPI support would be nice in general, though.

On Fri Dec 26 19:01:39 EST 2003, jmk@plan9.bell-labs.com wrote:
> On Fri Dec 26 18:06:45 EST 2003, mirtchov@cpsc.ucalgary.ca wrote:
> > quick question:
> >
> > will Plan 9 work with hyperthreading processors? will it recognize them as
> > dual-cpus?
> >
> > can you share experience with those?
> >
> > thanks: andrey
>
> Depends. In 2002 we bought a dual-processor Xeon machine with
> hyperthreading. After enabling the hyperthreading option in the
> BIOS setup Plan 9 detects and uses 4 processors. (I don't think
> it makes much, if any, improvement to the system's performance).
>
> Earlier this month we tried a new single-processor P4 system
> with hyperthreading. Even with the BIOS option set the BIOS MP
> table only contains a single processor entry. I haven't bothered
> yet to look to see if there's some BIOS problem or if there's
> some other way to take advantage of hyperthreading on this
> system.

Re: [9fans] fossil/venti

[okamoto]

> - plan 9 installation is getting complex. the easiest one i did was the
> first free version (just a standalone system with kfs). as features are
> added, it getting tougher.

I agree, and that is the reason why I posted my example for someone
who will have the same problem, too.
Complicated, because we have three kinds of file servers, and the last
one, which is expected to active now, is still in the progress.   If you just
want to try Plan 9, you can choose kfs version, because it's easier.

> - i think kfs file system is needed anyway atleast till fossil kernel is
> built and the config files are modified accordingly. it is not possible or
> easy to setup fossil/venti without kfs. am i right?

If you don't have working Plan 9 network, I think you need kfs system
first, from which you can fill the things into venti archive disk.☺

> (instead of
> installation-process-assisted fossil installation). 

I don't know this.  Is there?

> I used to feel I am just a end-user to Plan 9. 

Looks like so.☺

Kenji

Re: [9fans] fossil/venti

[andrey mirtchovski]

Here is the most straightforward way to get a fossil+venti system on a
single machine (not necessarily single disk):

	- install from cd/net, leaving a reasonable amount of space for venti
	  (the exact sizes are discussed in the wiki)

	- boot the brand new fossil system and configure the venti partitions as
	  explained in 'Configuring Venti' on the wiki

	- write the venti configuration to an arena partition (venti/conf)

	- set venti= in plan9 .ini to point to the  partition containing the venti
	  configuration

	- reboot, connect to the fossil console and do a 'snap -a', then
	  configure other parameters such as snaptimes. it's a good idea to
	  store the fossil and venti configurations offsite, just in case

As with most other Plan 9 things, the learning curve is very steep at the
beginning but, as many have found out, once realized, the system presents very
few surprises.

Having worked with venti and a crashy fossil for a while I've come to
realize that much more attention needs to be paid to venti since fossil
servers could be discarded freely, as long as the data in venti is intact.

andrey

Re: [9fans] fossil/venti

[vdharani]

>> - plan 9 installation is getting complex. the easiest one i did was
>> the first free version (just a standalone system with kfs). as
>> features are added, it getting tougher.
>
> I agree, and that is the reason why I posted my example for someone who
> will have the same problem, too.
yeah, i found the note you have posted only after sending my request.

> Complicated, because we have three kinds of file servers, and the last
> one, which is expected to active now, is still in the progress.   If
> you just want to try Plan 9, you can choose kfs version, because it's
> easier.
No, no. I have been using kfs (and venti) for a long time. I wanted to try
fossil with venti now.

>> - i think kfs file system is needed anyway atleast till fossil kernel
>> is built and the config files are modified accordingly. it is not
>> possible or easy to setup fossil/venti without kfs. am i right?

> If you don't have working Plan 9 network, I think you need kfs system
> first, from which you can fill the things into venti archive disk.☺
ah, now i get it. i was wondering how you have all tried fossil and
started discussing about it. i didnt know you have to use kfs initially or
some working plan 9 machine before fossil is in place.

this is what i tried to do:
the boot floppy asked whether i want kfs or fossil. i chose fossil. I was
thinking, instead of kfs, I can use fossil (with venti). looks like
fossil/venti can be up and running only after necessary steps are taken.

>> (instead of
>> installation-process-assisted fossil installation).
>
> I don't know this.  Is there?
yes.

Re: [9fans] hyperthreading

[ron minnich]

hyperthreading: don't waste your time.

hype-r-threading.

ron

Re: [9fans] hyperthreading

[Joel Salomon]

ron minnich said:
> hyperthreading: don't waste your time.
>
> hype-r-threading.
>
> ron

I'm to assume then that hyperthreading is distinct from having multiple
"cores" on a single die (what IBM is doing with some of their machines),
or is this the "hype" you are referring to? Multiple CPUs on one silicon
chip sounds cool to me.

--Joel

Re: [9fans] hyperthreading

[William Josephson]

On Sun, Dec 28, 2003 at 05:12:32PM -0500, Joel Salomon wrote:
> ron minnich said:
> > hyperthreading: don't waste your time.
> >
> > hype-r-threading.
>
> I'm to assume then that hyperthreading is distinct from having multiple
> "cores" on a single die (what IBM is doing with some of their machines),
> or is this the "hype" you are referring to? Multiple CPUs on one silicon
> chip sounds cool to me.

Look up the literature on SMT (symetric multithreading).
In the case of Intel, the two virtual CPUs do not have
the full set of resources available to a single real CPU.
The result is a modest improvement in performance for a
few specific workloads and has little or negative
performance impact on many others (cf.  the P4 core
generally -- it is a good example of microbenchmarking
gone awry in my experience).  I have found that with some
Unix variants, typically those with underwhelming
schedulers for interactive workloads, it helps a bit with
keeping the shell and windowing system zippy under load.

 -WJ

Re: [9fans] hyperthreading

[ron minnich]

On Sun, 28 Dec 2003, Joel Salomon wrote:

> I'm to assume then that hyperthreading is distinct from having multiple
> "cores" on a single die (what IBM is doing with some of their machines),

yes, it certainly is.

ron

Re: [9fans] hyperthreading

[Vasile Rotaru]

On Sun, 28 Dec 2003 17:12:48 +0100, ron minnich wrote:

> hyperthreading: don't waste your time.
> 
> hype-r-threading.
> 
> ron

Let's say if I understand it correctly.

I take a very simple example, say we need to fill a big array.

 case a: Single processor, single thread
   (the base case)

 case b: Single processor, two threads.
    (one fills the even, the other the odd positions).
     Slower, time lost in task switch.

 case c: Single HP processor two threads.
     A litle faster than in case b), since less time lost in task switch.

 case d) Two processors, two threads.
     Solid speed gain, almost two times faster.

 So I just can't think of a realistic scenario, where the speed of task
switching will be noticeable..

bas

Re: [9fans] hyperthreading

[David Presotto]

hyperthreading is another step down using as much of your resources as
possible.

With single threaded RISCs, we used instruction scheduling
to try to saturate the memory bus while computing.  This was pretty
straightforward and just required a good compiler backend.  Unfortnately,
it didn't work that well.  With only one thread, it was hard to always
find some memory accesses to do even with speculative execution.  You
could switch threads but that was expensive since you had to back out
much of the state of the current thread.

CISCs did pretty much the same thing with multiissue machines
and with Very Wide instructions.  Get enough stuff to do into the machine
so that at any one time the cpu elements and the memory bus were being
used at the greatest efficiency.

Hyperthreading is just another step down that path.  If you are executing
multiple threads simultaneously, there's some hope that you can find something
to use the memory bus for or utilize pipeline elements for while executing
the instructions.  It isn't so much an attempt to reduce context switch time
as it is an attempt to keep more options available to the processor's
internal scheduling.

Even with that, you're fighting a battle between using the chip area for
a truly hyperhtreaded processor and just sticking a separate processor core
on the chip.  Each thread increases the amount of synchronization (and
thus stalling) in the pipelines.  However, each extra thread uses up less
chip space than another core.  The point you give up and add another
processor (and syncronized write buffer) depends on the complexity
of the processor itself.  You're not likely to go to many threads
with an x86 and have it still look like an x86.  I believe
someone already pointed out that there really isn't that much
parallelism in the current hyperthreaded x86's.

In short, hyperthreading isn't a replacement for multiprocessors.  Its
just a way to build a single processor with greater (in some cases)
aggregate performance for the same (or not much greater) price.

Re: [9fans] hyperthreading

[ron minnich]

hyperthreading in concept is fine. It is just that so far, on Xeon, nobody
has seen a performance boost on almost any app, hence almost all the big
clusters I know of have turned it off and left it off.

ron