tlug: Re: multi-processor linux configuration ?

[Cliff Miller <cliff@example.com>]

Hi,

I got some email from a good friend of a good friend of mine, Andrej,
who is using Linux to do parallel and distributed processing to solve
real problems in biochemistry.

I've sent Andrej some of my opinions, but I thought I'd forward this
to the group and see what kind of opinions people come up with.  I've
pasted my comments to him at the end of the post.

I know that Andrej will be happy to get some input from you all on
this topic.

Cliff

> Date: Thu, 8 Oct 1998 22:54:36 -0400 (PDT)
> Cc: sali@example.com (Andrej Sali)
> Reply-To: sali@example.com
> From: sali@example.com
> X-Mailer: ELM [version 2.4 PL25]
> MIME-Version: 1.0
> Content-Type: text/plain; charset=US-ASCII
> Content-Transfer-Encoding: 7bit
> 
> Hi, Cliff,
> 
> We met at Stanford this summer, to talk about scientific linux applications ...
> As it happens, there is a real chance that we will be buying a multi-processor
> linux cluster, on the order of $50,000 to begin with. The system will be 
> used with Rok's CLUSTOR to run scalar jobs on the individual processors. They
> will require about 128-256MB of RAM each. Disk access via 100Mb/s NFS
> would be fine. The cluster should be reliable, even if it costs a little
> more. I would be most grateful for your opinion on the following 
> issues: What is the best setup for say 32 processors? From which manufacterer? 
> What version of Linux to use? Is it better to have 32 single processor 
> machines, or 16 dual processor machines, or that "cube" from a company
> in Salt Lake City? Is there a good web site that discusses various options?
> 
> Many thanks in advance! Andrej
> 
> P.S. We now have Turbolinux on 3 of our 4 PCs!
> 
> -- 
> --
> Andrej Sali, Assistant Professor
> The Rockefeller University, 1230 York Avenue, New York, NY 10021-6399
> tel +1 212 327 7550; lab +1 212 327 7206 ; fax +1 212 327 7540
> e-mail sali@example.com; http://guitar.rockefeller.edu
> 

================================================================
================================================================

Return-Path: <cliff>
Date: Wed, 14 Oct 1998 10:06:00 +0900
From: Cliff Miller <cliff@example.com>
To: sali@example.com
Cc: cliff@example.com
In-reply-to: <199810090254.CAA02781@example.com> (sali@example.com)
Subject: Re: multi-processor linux configuration ?

Hi Andrej,

It is great to hear from you! I've been on the road for about 10 days -- 
sorry it's taken till now to get back in touch.

> We met at Stanford this summer, to talk about scientific linux applications ...
> As it happens, there is a real chance that we will be buying a multi-processor
> linux cluster, on the order of $50,000 to begin with. The system will be 
> used with Rok's CLUSTOR to run scalar jobs on the individual processors. They
> will require about 128-256MB of RAM each. Disk access via 100Mb/s NFS
> would be fine. The cluster should be reliable, even if it costs a little
> more. I would be most grateful for your opinion on the following 
> issues: What is the best setup for say 32 processors? From which manufacterer? 
> What version of Linux to use? Is it better to have 32 single processor 
> machines, or 16 dual processor machines, or that "cube" from a company
> in Salt Lake City? Is there a good web site that discusses various options?

If communications speed is the most important, my feeling is that
you'll be best off using a machine where all processors are in one
"cube" machine, but that you'll be better off getting separate
machines if processing power is more important.  Separate machines
will probably be cheaper upfront than a "cube" machine, but you may
find that maintaining a large number of machines is rather
troublesome.

If you go with separate machines, I would recommend buying through VA
Research. You may pay slightly more, but these people have the most
experience in building Linux machines.

  www.varesearch.com

The cube machine guys are at:

  www.altatech.com

We would certainly be *very happy* if you decided to employ TurboLinux
on the machines and will be happy to provide you with "special" attention
when you need support. You may want to consider which Linux system to
use based upon the compiler. TurboLinux 2.0 uses gcc. TurboLinux 3.0
will use egcs, which handles C++ better and delivers faster code by as
much as 25%. 

I'll  pass these questions on to a few people who should be knowledgeable
about these matters and will give you some more feedback as it comes in.

> Many thanks in advance! Andrej
> 
> P.S. We now have Turbolinux on 3 of our 4 PCs!

That is terrific! Do you mind if we use this fact in our PR literature?

Cliff

=============
$B%/%j%U!&%_%i!<(B
$BBeI=<hDyLr<RD9(B
$B%Q%7%U%#%C%/%O%$%F%C%/3t<02q<R(B
Cliff Miller
Pacific HiTech, Home of TurboLinux
cliff@example.com

 
---------------------------------------------------------------
Next Nomikai: 20 November, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 12 December, 12:30 Tokyo Station Yaesu central gate
---------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp