Dear Ioana,
Your system is really big. My guess is that your system is about 2-3
orders of magnitude larger than the typical systems that other users
deal with regularly. In this case, indeed one should expect superb
scaling. You almost obtained perfect scaling!!! I would encourage you to
try the 512 CPUs. This is the best scaling data I've ever seen on
regular MD code (my guess is with PME?). My interpretation is that the
scaling reaches the peak at about 32-64 CPUs for typical 10-30K-atom
systems.
By the way, can we use your timing data as a benchmark? This is not a
joke because many other people have done the scaling this way. We feel a
little disadvantaged if we do not have a few real-world data to show
that AMBER can do just as well, if not better (thanks to the work of
Mike, Tom, and many others).
yong
***************
-----Original Message-----
From: Ioana Cozmuta [mailto:ioana_at_nas.nasa.gov]
Sent: Friday, November 15, 2002 10:22 PM
To: amber_at_heimdal.compchem.ucsf.edu
Subject: Re: question
Hello dr. Case,
Thank you for your message. I am not sure why you are saying that amber
does not scale well after 32/64 processors. How here are some data's
from a minimization run as function of the number of processors run an
origin 2000 machine with 512 processors.
#CPU's wall-time #minimization steps CPU-time
--------------------------------------------------------
16 8hours 375 127:38
64 2h:40min 500 169:11
128 1h:21min 500 170:49
So indeed the total CPU time is the same for the 64 and 128 processors
(which is what I think one would expect) however the real time run was
half if the number of processors was doubled. To me this is a pretty
good scaling.
I did not do the same test for dynamics but from the results of the
minimization procedure it seems worth increasing the maximum number of
processors allowed in amber.
Please let me know what you think. Thank you once again,
Ioana
************************************************************************
****
* Ioana Cozmuta *
*
* NASA-AMES Research Center * "Gravitation can not be held
responsible*
* Mail Stop 230-3 * for people falling in love"
*
* Moffet Field *
*
* phone: (650) 604-0993 * Albert
Einstein*
* fax: (650) 604-0350 *
(1879-1955) *
************************************************************************
****
On Fri, 15 Nov 2002, David A. Case wrote:
> On Fri, Nov 15, 2002, Ioana Cozmuta wrote:
> >
> > Maybe this is a naive question: why is the maximum number of
> > processors that one can use in amber limited to 128?
>
> Even on the most expensive computers, amber scales poorly after about
> 32 (or certainly 64) processors.
>
> > Is it trivial to modify this limit?
>
> Should not be hard....dac
>
> --
>
> ==================================================================
> David A. Case | e-mail: case_at_scripps.edu
> Dept. of Molecular Biology, TPC15 | fax: +1-858-784-8896
> The Scripps Research Institute | phone: +1-858-784-9768
> 10550 N. Torrey Pines Rd. | home page:
> La Jolla CA 92037 USA | http://www.scripps.edu/case
> ==================================================================
>
>
Received on Fri Nov 15 2002 - 20:40:56 PST
