From: "Sandeep Dutta"
Subject: [NMusers] NONMEM
Date: 1/26/06 2:00 PM
Hello: Does anyone in the NMUSERs group have any experience of
installing & using NONMEM on a Super Computer?
Sandeep Dutta
NONMEM
From: Juan Carlos Perin
Subject: Re: [NMusers] NONMEM
Date: 01/26/2006 01:32 PM
What do you mean by supercomputer? A cluster, with need to run batch jobs
in parallel? Or a large multiprocessor single computer system?
Juan Perin
Children's Hospital of Philadelphia
Bioinformatics Core
From: "Diane R Mould" drmould@attglobal.net
Subject: Re: [NMusers] NONMEM
Date: 01/26/2006 01:53 PM
Dear Sandeep
I assume you are referring to something like a CRAY? Are you aware of the fact
that Nonmem is single threaded and unless you make some substantial changes to the
source code, it cant take advantage of the parallel processing on a 'supercomputer'.
You would have to ask Mark Sale about his code changes, as I believe that he has
taking the concept of parallelizing Nonmem further than anybody else
Diane
From: Sandeep Dutta Sandeep.Dutta@abbott.com
Subject: Re: [NMusers] NONMEM
Date: Thursday, January 26, 2006 3:06 PM
Diane: Thanks. Yes I was talking about something like a CRAY. I didn't know that the
only way a supercomputer could reduce computation time was through parallel computing.
However, I would expect the higher processor speed would in of itself reduce run times.
I was aware that NONMEM is single threaded & had seen Mark Sale's presentation at
AAPS last year. But what I gathered from his talk was that it wasn't ready for prime-time as yet.
Sandeep
From: Sandeep Dutta Sandeep.Dutta@Abbott.com
Subject: Re: [NMusers] NONMEM
Date: Thu, 26 Jan 2006 13:38:42 -0600
A large multiprocessor in a single computer system! We already
have NONMEM running on a Linux cluster?
Sandeep Dutta
From: Sam Liao sliao@pharmaxresearch.com
Subject: Re: [NMusers] NONMEM
Date: Thu, 26 Jan 2006 14:59:09 -0500
Hi Sandeep:
Can you provide more details such as the operating system and fortran compiler used etc.?
Sam Liao
Pharmax Research
From: "Diane R Mould" drmould@attglobal.net
Subject: Re: [NMusers] NONMEM
Date: Thu, 26 Jan 2006 15:32:02 -0500
Sandeep
Processor speed is a primary factor in the run time, but the compiler, (including
the optimization used for the compile), the memory word and page size etc all play
a role in the overall computation time for a job. If for example you ran Nonmem on
the same PC with Compaq FORTRAN and Watcom FORTRAN (for example), you should see
noticeable differences in the run times. I ran benchmarks on several different
platforms about 4 years ago and found that some systems with apparently slower
speed processors actually ran the benchmark jobs a bit faster, so to some extent,
processor time is relative.
Diane
From: mark.e.sale@gsk.com
Subject: Re: [NMusers] NONMEM
Date: Thu, 26 Jan 2006 18:01:57 -0500
Sandeep,
You're correct that parallel NONMEM isn't ready for "prime time", or even late night. Don't
have a time line for it. The changes are extensive, and the current version is specific for
Windows, but will run on a cluster, SMP, LAN or even a grid (but not very well on a grid).
Anyone interested in parallel NONMEM, please let me know, I'll forward a list to Globomax and
perhaps other interested people. Stu actually had promised to collaborate on parallel NONMEM
after he finished v VI, we got him interested when we pointed out that he had been saying in
public for years that it couldn't be done. Unfortunately, Stu's work wasn't yet done.
Mark Sale M.D.
Global Director, Research Modeling and Simulation
GlaxoSmithKline
919-483-1808
Mobile
919-522-6668
From: "Xiao, Alan" alan_xiao@merck.com
Subject: Re: [NMusers] NONMEM
Date: Thu, 26 Jan 2006 21:44:05 -0500
Not much different from doing that on a PC. If your supercomputer is a UNIX system, for
example, just follow the NONMEM manual.
Alan
_______________________________________________________