[LUGA] Mit freundlicher Unterstützung von:
Linux New Media AG

Mail Thread Index

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: What are y'all doing with your smp machines? (fwd aus linux-smp)

---------- Forwarded message from owner-linux-smp-outgoing@vger.rutgers.edu on Wed, 28 Aug 1996 17:52:39 EDT -----------
Return-Path: <owner-linux-smp-outgoing@vger.rutgers.edu>
From: Christopher Neufeld <neufeld@physics.utoronto.ca>
Date: 	Wed, 28 Aug 1996 17:52:39 EDT
X-Mailer: Mail User's Shell (7.2.5 10/14/92)
To: linux-smp@vger.rutgers.edu
Subject: Re: What are y'all doing with your smp machines?
Sender: owner-linux-smp@vger.rutgers.edu
Precedence: bulk

   Since you asked....

   Well, I'm working for a government office which does policy change
impact forecasting. We have a large Monte-Carlo simulation which feeds
into a large calculator program. Hundreds of megabytes stream between
these programs, and our typical final output file is 750 MB long. From
there, it gets chewed upon by awk, sed, perl, sort, and whatever else,
usually in a pipeline of four or more programs, before producing
something of a size comprehensible to humans.
   The SMP has advantages for us in two areas. One, the Monte-Carlo
program has been parallelized using the bb_threads package (that's why I
wrote it, actually), so now runs much more quickly. Secondly, these long
pipelines can be sped up considerably, provided one is careful of the
ordering and tries to make sure that the bottleneck process is well
placed (for instance, a sort process is the ultimate bottleneck, nothing
gets past it until the previous pipeline elements have exited, and it
uses little CPU for affordable disk technology when acting on huge
files). In some cases, our disk access is speeded up by piping the output
of our program through gzip, and then using zcat to pipe the stream down
our long pipelines of analysis routines. In any case, the second
processor is rarely idle, and the improvement is quite noticeable.

   I gather our situation is unusual. I've got huge processing power on
my desktop, in the form of a 128 MB dual PPro 200 with 10GB of disk
space, and right next to it, on the other side of the telephone, a single
processor 70 MB Pentium-90 with 4.5GB of disk space. It is this latter
single-CPU machine which is the server, acting as our main NFS server,
printer server, tape server, and forwarding firewall, as well as the
machine I usually sit at to type on.
   We also have a dual Pentium-90 with 160 MB which we use for our model
analysis. While both model machines can and do serve their disks on NFS,
we don't rely on them for this most of the time because our calculator
program thrashes the disk continuously while cacheing information.

 Christopher Neufeld                   neufeld@physics.utoronto.ca
 Home page:  http://caliban.physics.utoronto.ca/neufeld/Intro.html
 "Don't edit reality for the sake of simplicity"

   _  | Peter J. Holzer             | If I were God, or better yet
|_|_) | Sysadmin WSR                | Linus, I would ...
| |   | hjp@wsr.ac.at               |     -- Bill Davidsen
__/   | http://wsrx.wsr.ac.at/~hjp/ |        (davidsen@tmr.com)

powered by LINUX the choice of a gnu generation
linux user group austria;
Letzte Änderung:
September 2010