Performance of Folding@Home and Distributed Folding cleints

User Ernie ernie at qbic.eis.net.au
Sat Nov 29 14:20:56 PST 2003


A question for the folks familiar with the Folding at Home protein folding
project.

I have been doing some performance testing of FreeBSD vs Linux in a couple
of the Distrbuted Computing projects particulary Folding at Home and
Distributed folding. The one that I am curios about is Folding at Home where
their is no specific FreeBSD client, instead it runs as a Linux client using
the linux_base-8 libraries from the ports collection. Under straight linux
redhat, CRUX, gentoo it takes about 13hours on a P4C 3.0Ghz for complete a
gromacs WU, the same hardware running either FreeBSD 4.9-RELEASE or
5.1-RELEASE is takes about 26hours to do the same task. The only think I
have noticed is that under Linux the client says it's using the SSE
instructions, where under FreeBSD and the linux_base-8 libraries it does not
say that. The cleint runs multi threaded under Linux or FreeBSD.

I have also noticed quite a performance difference, though not as extreme,
between the OS's in the distributed folding project where there is a native 
compile for FreeBSD. The benchmark suite says that FreeBSD is spending more 
time in sys (I/O) releated ativities than Linux but is still takes more
hours to get through set of 250 generations than Linux does, the significant
figure is the Foldtraj/ Usr time which seems to relate how long it will take
to complete a generation.

 Linux Benchmark P4C 3.0Ghz
 Summary
 -------
           Usr time  Sys time
           --------  --------
 Maketrj      2.670     0.710
 Foldtraj    28.930    11.000



 FreeBSD Benchmark P4 3.16GHz.
 Summary
 -------
           Usr time  Sys time
           --------  --------
 Maketrj      5.789     0.234
 Foldtraj    36.867     1.523


I am curious if there are any performace tweaks for FreeBSD, other than
client settings or nice levels, to get more performance out of 
either Folding at home or Distribute Folding clients?

I have asked this question on the respective forums, but the people there
are generally not FreeBSD gurus.

- Ernie.



More information about the freebsd-performance mailing list