SCIENTIFIC-LINUX-USERS Archives

November 2009

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Greg Kurtzer <[log in to unmask]>
Reply To:
Greg Kurtzer <[log in to unmask]>
Date:
Sat, 28 Nov 2009 22:25:26 -0800
Content-Type:
text/plain
Parts/Attachments:
text/plain (63 lines)
Hello.

In my experience I have seen performance differences which are
exaggerated by tightly coupled applications at scale (e.g. parallel
high performance clusters).

My tests were done implementing Perceus for cluster management, and I
tried to use as many of the same external library components as
possible (e.g. Intel Compilers, math libs, OFED, MPI, etc..). Because
I was using many of the same runtime components (and even shared
binaries) I didn't expect too large of a difference. I was wrong.
Performance deltas that are virtually unmeasurable on single systems
were compounded many times when running in tight unison at scale. At
most we found 30% performance deltas between RHEL and Centos when
running on just under 5k cores. Unfortunately time ran out so I wasn't
able to also test SL.

The reason for the difference that we found has to do primarily with
the build environment that Red Hat uses. RHEL is not built on RHEL but
rather a modified and highly optimized tool chain that is not
distributed. It would be reasonable to assume that Fedora is also
built using this environment.

BTW, this is one of the reasons that after founding Centos I continued
to also focus on a freely available high performance focused
distribution of Linux (Caos Linux) that is tuned and tested for high
performance (but lacks many of the packages as it is not general
purpose, which for those requirements I now run SL).

In the case of Fedora, you are also testing against a newer core
library set and kernel both of which may further skew the differences.

What benchmarks were you using to measure? What compilers did you use
and what was the hardware (if you don't mind me asking)?

Thanks for the insights and I am eager to hear if you are able to test with SL.

Greg



On Sat, Nov 28, 2009 at 9:11 AM, Lou Arnold <[log in to unmask]> wrote:
> I wanted to know if the SL distribution was designed specifically to be fast.
>
> I am running a CentOS V5.3 and found that it is noticeably slower than
> Fedora 11, to my great surprise.
>
> How does the SL distro compare for speed?
>
> I had planned to use the CERN or Fermi Lab version of SL if SL runs faster.
>
> --
> Lou.
>



-- 
Greg M. Kurtzer
Chief Technology Officer
HPC Systems Architect
Infiscale, Inc. - http://www.infiscale.com

ATOM RSS1 RSS2