Hi Joseph
> I have recently upgraded to amber9/mpich2. To my
> dismay, I found a significant hit in performance when
> using sander9, such that jobs running with sander8 are
> up to 5 times faster. I compiled mpich2 with Intel
> compiler8.1, and then compiled amber9 in parallel with
> the same compiler. I realize that something must have
> gone wrong, but it is not clear to me what it is.
> Any help on this matter would be greatly appreciated.
> I am running on a cluster with Intel xeon processors
> and Red Hat Linux Entreprise 3.0.
This is a tough one to help out on due to all the variables. However, a few
questions. Are you using sander or pmemd? What type of simulation are you
running? I you are doing periodic boundary PME or Generalized Born implicit
solvent you should be using PMEMD as you will get better performance.
Secondly have you compiled up amber 8 with mpich2? Does this show slower
performance that with mpich v1?
What about single processor speed, without mpich. Run a few of the
benchmarks in $AMBERHOME/benchmarks as single processor with amber 8 and
amber 9. Amber 9 should come out faster than amber 8. If it does then the
problem is definately with your mpich2 installation. If it comes out much
slower then you possibly have an issue with your compilers / libraries you
are linking to etc.
Anyway, I suspect it is a problem with the mpich2 installation. What is the
interconnect on your cluster? Be careful that you haven't done something
like compile up mpich2 to use the ethernet interconnect instead of say
myrinet or infiniband. Run some of the mpich2 test cases that come with it
to check.
The essential thing here is, like with any scientific experiment, to reduce
the problem down to a set of independent variables that you can test one at
a time.
All the best
Ross
/\
\/
|\oss Walker
| HPC Consultant and Staff Scientist |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk | PGP Key available on request |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun May 14 2006 - 06:07:11 PDT
