CCL:parallel quantum solutions (fwd)

[Eugen Leitl]

---------- Forwarded message ----------
Date: Tue, 09 Apr 2002 10:07:54 -0400
From: David J Giesen <david.giesen at kodak.com>
To: Dr. Bill Davis <bdavis at UTB.edu>, CHEMISTRY at ccl.net
Subject: CCL:parallel quantum solutions

Bill -

This is a long post, but I think vendors that do excellent jobs should
get pats on the back.  I am not a PQS employee nor do I receive
new-customer kick-backs.  Those not interested in a review of PQS
products can hit delete now.

I've been a very satisfied PQS customer since Aug. 2000.  We purchased
an 8-processor Linux cluster at that point, and several months later we
were so pleased we bought a second.  At the end of last year, we
contracted with them to build a 34-processor cluster for general
computational (not just chemistry) use.  

Hardware performance : The setup they use works well for running serial
or parallel codes, and I have PQS, Jaguar and Gaussian (using LINDA)
running in parallel on them.  Based on timings against other
machines/platforms, the PQS machines perform as well as could be
expected.  Our 1.2 GHz athlon PQS machine runs G98 slightly (~10%)
faster than the latest-and-greatest-just-off-the-design-sheet Sun
hardware, and 2-3 times faster than our SGI 194 MHz R10000.  It is ~10%
slower than a 1.5GHz P4.  Both the Athlon and P4 machine used PIII
optimized blas libraries...

Software performance : the PQS software 'is what it is'.  If you are
interested in mainly HF, MP2 and DFT computations, it is very good.  You
can see its capabilities on their website.  Speed-wise, it runs faster
than other codes I use, although it is not faster than Jaguar's
pseudo-spectral methods.  The geometry optimizer is rock solid
dependable as one would expect from code by Pulay and Baker.  PQS uses
PVM for parallel execution.  Without getting into a debate about
parallel paradigms, I'll say simply this: in our hands, I have never had
a PVM job die because of inter-process communication problems while
MPICH/MPI is very flaky and tends to die on about 10-25% of chemistry
jobs (even more for systems using automount) independent of linux, sun
or SGI.  Because PQS uses PVM to set up the parallel system only once
per job, there is less parallel overhead using PQS than with other codes
that set up LINDA parallel systems at every SCF and geometry
optimization step - although for large jobs, these both essentially go
to zero.

Support : PQS is a small company, and the support shows it.  They have
absolutely bent over backwards each time we have had an issue, and
dealing with them is always a pleasure.  We have not had a hardware or
PQS software issue that they haven't resolved to our satisfaction.  In
fairness, you can't expect a 24-hour help line or technicians in suits
to fly in and fix problems.  You should be aware that they are not in
the business of selling/supporting Linux or Gnu software, so some
problems you have on your machine if you veer off the PQS path might be
technically out of their scope.  However, in my experience, they make
every honest effort to solve those as well (and usually do).  Every
machine they shipped us has been stress-tested by an expert for a number
of days before they are delivered.

Ease of use : The machines come setup to run the PQS chemistry code out
of the box.  If you are planning on running one PQS parallel job at a
time across the whole cluster or multiple serial jobs, the included DQS
(not associated with PQS) queuing system works OK.  Running multiple
parallel codes/jobs on the same cluster through the queue does not work
well. Running other codes in parallel through the queue takes some hard
work.  Setting up other parallel codes also takes some work.  This is
not really a function of PQS, however, and you'll find this is true no
matter what machine you get.

Disclaimer : This e-mail does not in any way imply an 'official Kodak'
stance, it is merely the personal opinion of a Kodak employee who uses
PQS products at work.

Dave

Dr. Bill Davis wrote:
> 
> Hi!
> 
> Does anyone have any experience with the PQS hardware/software
> combination, more specifically the QS4-1800S?  Any comments on ease of
> use, support and any other important points would be greatly
> appreciated...Thanks!
> 
> Bill
> 
> 
> --
> **********************************
> Dr. William M. Davis
> Assistant Professor of Chemistry/
> Phi Theta Kappa Advisor
> Dept. of Chemistry and Environmental Science
> University of Texas at Brownsville
> 80 Fort Brown
> Brownsville, TX 78520
> Phone: (956) 574-6646
> Fax: (956) 574-6692
> WWW: unix.utb.edu/~bdavis
> **********************************
> 
> 

-- 
Dr. David J. Giesen
Eastman Kodak Company                           david.giesen at kodak.com
2/83/RL MC 02216                                (ph) 1-585-58(8-0480)
Rochester, NY 14650                             (fax)1-585-588-1839


-= This is automatically added to each message by mailing script =-
CHEMISTRY at ccl.net -- To Everybody  | CHEMISTRY-REQUEST at ccl.net -- To Admins
MAILSERV at ccl.net -- HELP CHEMISTRY or HELP SEARCH
CHEMISTRY-SEARCH at ccl.net -- archive search    |    Gopher: gopher.ccl.net 70
Ftp: ftp.ccl.net  |  WWW: http://www.ccl.net/chemistry/   | Jan: jkl at osc.edu