[Bioclusters] Call for information.

[Goran Ceric]

> 
> Hmm... if that's the case, how would you rate quad-P3Xeon (1MB and 2MB
> cache)?
> 

Let me intercept this. We have one dual P4 1.7 GHz machine and its 
performance is horrible compared to dual 1.4 GHz Tualatin machines. We also 
have a couple of 4-way 700 MHz P3 Xeon/2MB machine and its single CPU 
performs at 50% of the above mentioned Tualatins (clock speed ?). I don't 
think that in our case size of L2 cache plays that big of a role. Most of 
the stuff we run seems to be I/O and sometimes memory-bound anyway. We do 
have some strictly CPU-bound programs. And there's of course all these 
little programs that are not threaded so a large SMP box wouldn't help that 
much anyway. What I'd like to hear myself is people's experiences with 
Itanium-based systems. 

Besides the hardware, the OS itself plays a big role. I tested a lots of 
different Linux kernels running different benchmarks on these machines and 
wouldn't suggest anyone using anything less than 2.4.17 (A. Cox-patched 
kernels always seem to be better). I didn't have good luck with Redhat-
supplied 2.4.7-10smp and 2.4.9-21smp kernels (mostly NFS, Reiserfs, and >2 
GB files related things). I did have success with Rawhide 2.4.17-0.16smp 
and up kernels even though they are considered experimental (that's if you 
don't want to compile them yourself and you're using Redhat). Again, this 
is based only on my experience. 

Goran Ceric
System Administrator
Washington University St. Louis
Department of Genetics, Eddy Lab
goran@genetics.wustl.edu
http://www.genetics.wustl.edu/eddy