Re: [9fans] nice hardware for a cpu server

Re: [9fans] nice hardware for a cpu server

[YAMANASHI Takeshi]

On Tue Feb  3 10:54:26 JST 2004, Charles Forsyth wrote:
> to be more precise, the answer to the good question
> 	If you were plowing a field what would you rather use, 2 strong oxen or 1024 chickens? -Seymour Cray
> seems to be ``well of course we'll use 1024+ strong(-ish) oxen''

Adding to that, in grid people, `share' means ``aren't there
any idle oxen down on which I can get laid my burden?   My
oxen are full already.''

Of course, everybody's oxen are full and there is no `share'
in them resulting the failure of inter-organization grid.

There can be a intra-organization grid.  But what it differs
from a cluster?
--

Re: [9fans] nice hardware for a cpu server

[andrey mirtchovski]

> There can be a intra-organization grid.  But what it differs
> from a cluster?

A cluster generally assumes almost identical hardware, software and
capabilities.  Intra-organizational resource sharing is considered an
example of Meta/Distributed Computing (metacomputing was the buzzword
of choice after Distributed Computing's demise and before the rise of
Grids), while Grid Computing is resource sharing across administrative
domains...

As for oxen vs.  chickens -- that's a pretty tough one to decide
considering that today's commodity chicken can pull more than the
average ox.  Or at least that's the impression one gets out of looking
at the SPEC site...  Maybe chicken vs fleas is better, throwing in a
side note about the extinction of the Ox and how it affected the
world...

Andrey

PS: I'm too young to have cared about computers much when War Games
came out, but I've seen the pictures of the CM5 at LANL -- an Ox
composed of 64000 chickens :)

Re: [9fans] nice hardware for a cpu server

[Charles Forsyth]

>>As for oxen vs.  chickens -- that's a pretty tough one to decide
>>considering that today's commodity chicken can pull more than the

i was being kind and considering the 3GHz Pentium-4 PC that's apparently
required to run documents and spreadsheets not to mention games to be an ox.
certainly i find it acts like one with Windows, and Linux isn't far behind.

Re: [9fans] nice hardware for a cpu server

[ron minnich]

On Tue, 3 Feb 2004, YAMANASHI Takeshi wrote:

> Adding to that, in grid people, `share' means ``aren't there
> any idle oxen down on which I can get laid my burden?   My
> oxen are full already.''

The first problem is, that going back to the 70s, grids have never worked.
Check out rsexec from early 70s and tell me how Globus is any better. For
the decades of the 70s, 80s, 90s, and now, you can find some term that
corresponds to the term 'grid'. And you'll find that people had great
dreams and they all went down in flames, and the terms ended up getting
used by marketing to mean something other than they were supposed to.

This is happening now with grids -- Oracle in its new ads is calling
clusters grids!

The one ray of light in this mess is Plan 9. I'm convinced (probably
wrongly, as usual, but ...) that Plan 9 has really got what it takes to
really make grids a reality.

ron

Re: [9fans] nice hardware for a cpu server

[Martin C.Atkins]

On Mon, 2 Feb 2004 20:16:13 -0700 (MST) ron minnich <rminnich@lanl.gov> wrote:
> The one ray of light in this mess is Plan 9. I'm convinced (probably
> wrongly, as usual, but ...) that Plan 9 has really got what it takes to
> really make grids a reality.

Call me stupid, but I've yet to understand what a 'grid' really has
to offer me. Then again, I'm not in your business :-) !

Martin

--
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] nice hardware for a cpu server

[ron minnich]

On Tue, 3 Feb 2004, Martin C.Atkins wrote:

> Call me stupid, but I've yet to understand what a 'grid' really has
> to offer me. Then again, I'm not in your business :-) !
>

nobody's stupid, just things look different from different places.

Lots of use has been made of "grids" or whatever you want to call them for
10-15 years now. What's really, really weird is that the examples come and
go, and everyone forgets them. I think I'd better start a section on my
web page.

ron

Re: [9fans] nice hardware for a cpu server

[Martin C.Atkins]

On Mon, 2 Feb 2004 21:34:09 -0700 (MST) ron minnich <rminnich@lanl.gov> wrote:
> On Tue, 3 Feb 2004, Martin C.Atkins wrote:
> > Call me stupid, but I've yet to understand what a 'grid' really has
> > to offer me. Then again, I'm not in your business :-) !
>
> nobody's stupid, just things look different from different places.

Thanks for your broadmindedness!

> Lots of use has been made of "grids" or whatever you want to call them for
> 10-15 years now. What's really, really weird is that the examples come and

I'm sure that's true. And I'm not saying that grids are not extremely
useful for super-computer-type computing. Just that it seems pretty
irrelevant (in contrast to some of the hype?) in other problem
domains. Hence...

> go, and everyone forgets them. I think I'd better start a section on my
> web page.

I'll look forward to learning more, and particularly look out for
counter-examples to my statement above (the one about irrelevance!).

Thanks!

Martin

--
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] nice hardware for a cpu server

[ron minnich]

On Tue, 3 Feb 2004, Martin C.Atkins wrote:

> I'm sure that's true. And I'm not saying that grids are not extremely
> useful for super-computer-type computing. Just that it seems pretty
> irrelevant (in contrast to some of the hype?) in other problem
> domains. Hence...

such as ...

- test vector generation for new chips
- semiconductor simulation
- financial modeling
- distributed  auto-place and route for FPGAs
- jet engine simulation

are these supercomputer apps? They're all very old uses of "grids", dating
back 12-15 years, 3 of them used Condor, one of them I ran (FPGAs).

ron

Re: [9fans] nice hardware for a cpu server

[Taj Khattra]

On Mon, Feb 02, 2004 at 08:16:13PM -0700, ron minnich wrote:
> The one ray of light in this mess is Plan 9. I'm convinced (probably
> wrongly, as usual, but ...) that Plan 9 has really got what it takes to
> really make grids a reality.

vitanuova's inferno based grid stuff (www.vitanuova.com/solutions/grid/)
looks neat (well it's neat to read about in any case, compared to the
globus gooware).

-taj

Re: [9fans] nice hardware for a cpu server

[andrey mirtchovski]

> vitanuova's inferno based grid stuff (www.vitanuova.com/solutions/grid/)
> looks neat (well it's neat to read about in any case, compared to the
> globus gooware).

VitaNuova has a great selling point for Inferno Grids which is often
used to belittle Plan 9's viability in this area -- their stuff can
easily be integrated with other operating systems.

All I'm hearing around here is "The world isn't going to switch and
run only Plan 9 you know".  Well, Inferno can deliver interoperability
now, which is better than saying "what the world needs to do is
implement 9p" :)

I'm waiting for publications to come out describing Inferno Grids in
better detail ...

andrey

Re: [9fans] nice hardware for a cpu server

[Bruce Ellis]

there's a pretty decent grid going on here.
not 4e; but inferno (and plan9) are simply
designed to utilize distributed resources.
my webserver runs on my linksys but all
of the resources are remote, cluttering my
pad enourmously.

the linksys imports '#C' from a p9 machine
so it can run presto's whois to Hostile the
hackers.  the big files are on various machines.
the cache is on a machine with a lot of memory.
it's all good.

product is "Mined Dealed Delivered".
yeah i did the pjw and got into trading.

brucee
----- Original Message -----
From: "andrey mirtchovski" <mirtchov@cpsc.ucalgary.ca>
To: <9fans@cse.psu.edu>
Sent: Tuesday, February 03, 2004 4:45 PM
Subject: Re: [9fans] nice hardware for a cpu server


...
> I'm waiting for publications to come out describing Inferno Grids in
> better detail ...
>
> andrey

Re: [9fans] nice hardware for a cpu server

[C H Forsyth]

>>of the resources are remote, cluttering my
>>pad enourmously.

sounds like my loft.

Re: [9fans] nice hardware for a cpu server

[C H Forsyth]

>>VitaNuova has a great selling point for Inferno Grids which is often
>>used to belittle Plan 9's viability in this area -- their stuff can
>>easily be integrated with other operating systems.

the latter part is true, but i don't think i'd use it to belittle Plan
9's use in certain grid work.  as brucee described, one can put them
in useful combination.  i still use them both.
(that's not to say that i haven't got plans.)

Re: [9fans] nice hardware for a cpu server

[Jack Johnson]

On Feb 3, 2004, at 7:51 AM, C H Forsyth wrote:
> in useful combination.  i still use them both.
> (that's not to say that i haven't got plans.)

Does that mean we can expect a Ghostscript port to Inferno shortly?

;)

-Jack

Re: [9fans] nice hardware for a cpu server

[Bruce Ellis]

just import '#C' from a p9 hosted inferno and use the os command.

----- Original Message -----
From: "Jack Johnson" <fragment@nas.com>
To: <9fans@cse.psu.edu>
Sent: Wednesday, February 04, 2004 2:53 PM
Subject: Re: [9fans] nice hardware for a cpu server


> On Feb 3, 2004, at 7:51 AM, C H Forsyth wrote:
> > in useful combination.  i still use them both.
> > (that's not to say that i haven't got plans.)
>
> Does that mean we can expect a Ghostscript port to Inferno shortly?
>
> ;)
>
> -Jack

Re: [9fans] nice hardware for a cpu server

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 157 bytes --]

since Ghostscript seemed to cross the Styx some time ago,
i suppose it must end up in Inferno sooner or later.
Circle 3 or Circle 4, that's the question.

[-- Attachment #2: Type: message/rfc822, Size: 2360 bytes --]

From: Jack Johnson <fragment@nas.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] nice hardware for a cpu server
Date: Tue, 3 Feb 2004 19:53:34 -0800
Message-ID: <B4A001D0-56C5-11D8-A074-000A95E29604@nas.com>

On Feb 3, 2004, at 7:51 AM, C H Forsyth wrote:
> in useful combination.  i still use them both.
> (that's not to say that i haven't got plans.)

Does that mean we can expect a Ghostscript port to Inferno shortly?

;)

-Jack

Re: [9fans] nice hardware for a cpu server

[Jack Johnson]

> On Feb 3, 2004, at 7:51 AM, C H Forsyth wrote:
>> in useful combination.  i still use them both.
>> (that's not to say that i haven't got plans.)

Just slightly more seriously, which apps are keeping you from going
entirely Inferno?

-Jack

P.S.
I'm a little slow on the uptake, so very belatedly thank you for
posting the source to the BeBox port!  Now if I can just finish
cleaning the garage....

Re: [9fans] nice hardware for a cpu server

[Bruce Ellis]

all i need is a hosted inferno on any os which has functionality
or resources that are either impractical to port or proprietary.
why would i write troff in limbo when i can use 'os troff'?
rob said in very early inferno days something along the lines
of "don't just rewrite everything in limbo, it's a different world".
the ghostscript example has always been there as a one-liner.

brucee
----- Original Message -----
From: "Jack Johnson" <fragment@nas.com>
To: <9fans@cse.psu.edu>
Sent: Thursday, February 05, 2004 6:21 PM
Subject: Re: [9fans] nice hardware for a cpu server


> > On Feb 3, 2004, at 7:51 AM, C H Forsyth wrote:
> >> in useful combination.  i still use them both.
> >> (that's not to say that i haven't got plans.)
>
> Just slightly more seriously, which apps are keeping you from going
> entirely Inferno?

Re: [9fans] nice hardware for a cpu server

[Charles Forsyth]

>>of "don't just rewrite everything in limbo, it's a different world".

yes, it's sensible to avoid deja vu.
i've seen all the same things too many times now as it is.

Re: [9fans] nice hardware for a cpu server

[boyd, rounin]

> yes, it's sensible to avoid deja vu.

and déjà vu ;)

Re: [9fans] nice hardware for a cpu server

[Jeff Sickel]

Charles Forsyth <forsyth@terzarima.net> wrote:

> yes, it's sensible to avoid deja vu.

DjVu on the other hand was a very nice wavelet based image system...

If only the world wasn't going towards vector graphics as much as it is these
days.

jas
___
remove the spamisham.

Re: [9fans] nice hardware for a cpu server

[rog]

> DjVu on the other hand was a very nice wavelet based image system...

what are wavelets?

Re: [9fans] nice hardware for a cpu server (wavelets)

[Steve Simon]

Mmm, difficult to explain without a piece of paper.

jpag takes DCTs of the image, quantizes the coefficenst and encodes these as
the jpeg image (roughly).

Wavelet compression (now in Jpeg2000 standard) uses an arbitary waveform
(arbitary but fixed and well known) to peform the transform the image.
The idea is that the distribution of visually important coefficents is
easier to qualify in wavelat space rather than in DCT (which is near 2D
frequency) space.

Wavelets have many interesting properties
image compression is just one application of course.

Loads more info at http://www.wavelet.org
though you may be amused to know that this points to the same machine as....
plan9.bell-labs.com; they are clever people there.

-Steve

----- Original Message -----
From: <rog@vitanuova.com>
To: <9fans@cse.psu.edu>
Sent: Friday, February 06, 2004 9:08 PM
Subject: Re: [9fans] nice hardware for a cpu server


> > DjVu on the other hand was a very nice wavelet based image system...
>
> what are wavelets?
>
>

Re: [9fans] nice hardware for a cpu server

[Jeff Sickel]

rog@vitanuova.com wrote:
> > DjVu on the other hand was a very nice wavelet based image system...

> what are wavelets?

In this case the wavelets are used for bitmap image compression algorithms.

Unfortunately, ASCII doesn't exactly produce the equation correctly:

http://documents.wolfram.com/applications/wavelet/index2.html

But in the exact reference I was mentioning, http://djvu.sourceforge.net/
would be your best bet.  You could also look at http://www.wavelet.org/ though
that wouldn't exactly give you the most discrete definition.

Wavelet algorithms have also shown up in JPEG2000, but I'm not familiar
enough with that spec to really comment.

Unfortunately, the djvu base library is implemented in C++ which would put
the Plan 9 C compiler out of the running.  But nudging the port into a C
version would be a nice thing as I've had problems with various gcc versions
in the past.

jas
___
remove spamisham.

Re: [9fans] nice hardware for a cpu server

[vdharani]

> all i need is a hosted inferno on any os which has functionality
> or resources that are either impractical to port or proprietary.
> why would i write troff in limbo when i can use 'os troff'?
> rob said in very early inferno days something along the lines
> of "don't just rewrite everything in limbo, it's a different world".
> the ghostscript example has always been there as a one-liner.

but wouldnt it be nice having everything in limbo and use it in all
platforms that run inferno? in the troff example, yes, you can do it. but
imagine you have troff in limbo. the benefits are definitely much more.

i feel limbo is a simple, nice language with which one can write reliable
apps. i am really surprised why people (including creators, people at the
lab, and rest of the 9fans) tend to ignore limbo.

i recently saw something like "370 Applications in Java" in a magazine
cover and wondered why limbo is so ignored even by us (9fans, inferno
fans). what is the actual problem? is it to do with licensing? or
something really wrong with limbo?

i understand existing apps that are huge and written in C may be difficult
to port to limbo. but why cant we write atleast new apps in limbo? if
licensing is an issue, cant vita nuova fix it or relax it for the sake of
saving both plan 9 and inferno?

regards
dharani

Re: [9fans] nice hardware for a cpu server

[Bruce Ellis]

> but wouldn't it be nice having everything in limbo and use it in all
> platforms that run inferno? in the troff example, yes, you can do it. but
> imagine you have troff in limbo. the benefits are definitely much more.

well let's go for a more extreme (real world) example.  why would i
write a multi band parametric equalizer and a small room reverb
simulator in limbo when i can simply import that functionality from a
proprietery device with lots of dsps?

> i understand existing apps that are huge and written in C may be difficult
> to port to limbo. but why cant we write atleast new apps in limbo?

every new app i write is in limbo or mash.

brucee

Re: [9fans] nice hardware for a cpu server

[a]

sure, there's tons of stuff you wouldn't want to rewrite just because
there's a new system to target it for (unless your from certain GNU
projects). but at the same time, i often don't have a plan9/unix box
around to "os" to to run troff or produce PDF files for me. not to
mention the fact that some - many - things don't lend themselves well
to that sort of "fire and forget" operation. while we needn't port
ghostscript, per se, it would be nice to be able to produce PS or
PDF from inferno without the extra box.

now, yes, brucee's right: if you've got a box that does what you
want already, and you have access to that box whenever you want to
do that, figure out how to export the resource rather than porting
it. we've had styx interfaces to OCR systems on remote unix boxes
and TTS and ASR systems on boards running an embedded unix. if we
could have had a equally-performant version of these written in
limbo directly, we'd have been happier, but we didn't really care
much: we were interested in building with them as parts, and once
we had the interface to the parts, their internal construction was
a matter of curiosity and taste more than anything else.

besides, the more time spent rewriting existing stuff is less time
spent writing new stuff. you can do in limbo almost all of what
you can do in other languages, but you can also do things in limbo
that are impossible (or nearly so) in other languages. i know where
i'd rather spend my time.
ア

Re: [9fans] nice hardware for a cpu server

[boyd, rounin]

> besides, the more time spent rewriting existing stuff is less time
> spent writing new stuff. you can do in limbo almost all of what
> you can do in other languages, but you can also do things in limbo
> that are impossible (or nearly so) in other languages. i know where
> i'd rather spend my time.

exactly. i know where i'm trying to spend mine ...

Re: [9fans] nice hardware for a cpu server

[Caerwyn Jones]

Bruce Ellis wrote:
> all i need is a hosted inferno ...

really, anything that can handle styx/9P2000.
The ideas in the Styx Architecture paper have only recently began to
sink in for me.

Re: [9fans] nice hardware for a cpu server

[Bruce Ellis]

wow, you are a lot sharper than some fresh folk from the inferno BU
who nevere got the little picture, let alone the big one.

brucee
----- Original Message -----
From: "Caerwyn Jones" <caerwyn@comcast.net>
To: <9fans@cse.psu.edu>
Sent: Friday, February 06, 2004 8:55 PM
Subject: Re: [9fans] nice hardware for a cpu server


> Bruce Ellis wrote:
> > all i need is a hosted inferno ...
>
> really, anything that can handle styx/9P2000.
> The ideas in the Styx Architecture paper have only recently began to
> sink in for me.

Re: [9fans] nice hardware for a cpu server

[a]

exactly. see also the styx-on-a-brick project/demo VN did
getting a styx server running on the lego mindstorm kit.
sure, it's a toy, but really interesting. it exported a
namespace with sensor and motor files. you then build
other apps on top to make use of the resources. for
perversity, while doing a demo, i had the IR link to
the lego box (that's serial+IR links) in /chan on box A,
exported (over TCP/IP on ethernet) to box B who mounted
it into /mnt. box B exported that /mnt namespace to box
C (again, TCP/IP and ethernet) which ran a program that
accepted writes on /mnt/clock and turned them into sets
of commands into the /mnt namespace that would make the
clock show that time. then box C exported that file to
box D (in this case a iPaq, connected over WiFi) which
ran a GUI program where you could drag the hands or a
clock around or click somewhere and it would write the
time into /chan/clock. as far as every bit was concerned
the earlier bits may as well have been running remotely.
it worked wonderfully, and showing people how easy it
was to plug together (heck, it takes longer to
explain!) was a killer demo.
ア

Re: [9fans] nice hardware for a cpu server

[ron minnich]

On Mon, 2 Feb 2004, andrey mirtchovski wrote:
> All I'm hearing around here is "The world isn't going to switch and
> run only Plan 9 you know".

"The world isn't going to switch and run Unix"
"The world will never use Linux"
"The world will never use PCs"

Take a year, take a new idea, you can find people who will dump on it.

ron

Re: [9fans] nice hardware for a cpu server

[Jack Johnson]

On Feb 2, 2004, at 6:20 PM, YAMANASHI Takeshi wrote:
> Of course, everybody's oxen are full and there is no `share'
> in them resulting the failure of inter-organization grid.
>
> There can be a intra-organization grid.  But what it differs
> from a cluster?

If latency isn't an issue (ha ha), geographically disparate
organization sharing would be interesting.  Borrow cycles from an
organization eight timezones away.

-Jack

Re: [9fans] nice hardware for a cpu server

[ron minnich]

On Mon, 2 Feb 2004, Jack Johnson wrote:

> If latency isn't an issue (ha ha), geographically disparate
> organization sharing would be interesting.  Borrow cycles from an
> organization eight timezones away.
>

It's been done, and history forgotten. In the early 90s, IBM Watson ran a
global Condor flock, Tokyo->Zurich. You could watch activity follow the
terminator (day/night) line.

ron

Re: [9fans] nice hardware for a cpu server

[boyd, rounin]

> It's been done, and history forgotten.

yeah, late '90s the bank i was contracted to had all their mail/web
services run/managed and firewalled out of paris.  now if they'd
just _told_ me that before i got sent out to do sendmail hackery.

but, the shell and an awk script smashed 10k addresses through
it, so it looked ok.  sort of 'programing the inputs', but really
verifying the outputs based on what the old logs did.

Re: [9fans] nice hardware for a cpu server

[vdharani]

> On Feb 2, 2004, at 6:20 PM, YAMANASHI Takeshi wrote:
>> Of course, everybody's oxen are full and there is no `share'
>> in them resulting the failure of inter-organization grid.
>>
>> There can be a intra-organization grid.  But what it differs
>> from a cluster?
>
> If latency isn't an issue (ha ha), geographically disparate
> organization sharing would be interesting.  Borrow cycles from an
> organization eight timezones away.
may be, in a way, recent viruses are good examples? :-)

regards
dharani

Re: [9fans] nice hardware for a cpu server

[vdharani]

>> On Feb 2, 2004, at 6:20 PM, YAMANASHI Takeshi wrote:
>>> Of course, everybody's oxen are full and there is no `share'
>>> in them resulting the failure of inter-organization grid.
>>>
>>> There can be a intra-organization grid.  But what it differs
>>> from a cluster?
>>
>> If latency isn't an issue (ha ha), geographically disparate
>> organization sharing would be interesting.  Borrow cycles from an
>> organization eight timezones away.
> may be, in a way, recent viruses are good examples? :-)
btw, i meant software viruses.

>
> regards
> dharani