----- Original Message -----
From: "Alan Jackson" <[EMAIL PROTECTED]>
To: "Richard Anderson" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
Sent: Monday, September 17, 2001 4:23 PM
Subject: Re: BP adds to Open Source


>
> I'm also not sure your statement about Fortran vs C is correct. We've done
> benchmarks, and for hard core I/O + CPU numerics, Fortran is still hard to
> beat. And we run ours on Linux (the 1024 node IBM cluster). Porting to
Linux
> was amazingly easy.
>
If you have some relevant, non-proprietary benchmarks for the performance of
seismic code in Fortran on Linux vs. C on Linux using an open-source
compiler, let's see them.  I was unable to find an open source Fortran
compiler for Linux - which one are you using?

Since the GCC compiler has been in use on Linux for about a decade, I would
expect that it is pretty bug-free and well-tuned.

My claim is partly based on my experiences in the 80s with the kind of code
that grizzled oil-industry Fortran programmers produce vs. the kind of code
relatively young academic C programmers produce.  Given identical algorithms
and coding styles, any difference in the code performance would most likely
be due to a difference in the compilers.  Again, I'd expect GCC to do a
little better than a relatively immature Fortran compiler.

> We have an in-house system that is so well-tuned that when we ran it on
> Apollo workstations (remember those?), it overloaded the power supplies
> and caused them to arc out and fail. They never expected anyone would
> drive the I/O, CPU, and Bus at 100% all at once.

Great, but I don't see how this is relevant to the performance of the system
on Linux.

Is it still possible to get replacement parts for Apollos?

Richard Anderson, Ph.D
www.raycosoft.com
[EMAIL PROTECTED]
Seattle, WA


_______________________________________________
http://linux.nf -- [EMAIL PROTECTED]
Archives, Subscribe, Unsubscribe, Digest, Etc 
->http://linux.nf/mailman/listinfo/linux-users

Reply via email to