AMBER Archive (2000)

Subject: Re: parallel problem ?

From: Alexandre Gillet (gillet_at_glycine.jouy.inra.fr)
Date: Mon Mar 06 2000 - 16:30:56 CST

here is more information about the parallele machine I am using;

machine: 4 biprocessor PC pentium III 450 MHz + 1 monoprocessor PC
pentium III 500MHz ( Master processor)
Os : Red Hat Linux 6.0 Kernel 2.2.5-15
MPI version : MPI 1.2 under LAM 6.2b

> What bandwidth is the connection between
> processors?

100 Mbytes/ sec.

> How fast is it if you run with 2, 4, and 8 processors?

for 1 processor ( using the master PC) cpu = 2.26 result obtain after 2
hours 25. ( non parallel amber used )
for 3 processors cpu time = 0.98 result obtain after 1 hour 45.
for 5 processors cpu time = 0.63 result obtain after 2 hours 17.
for 7 processors cpu time = 0.47 result obtain after 2 hours 20.

>
> How many atoms?

I am using 13552 atoms.

Thanks you for your help.

> Dear amber users,
>
> I am running amber 5 on a parallel machine using MPI.
> The program run well but the time I get are quiet long.
> On 9 processors the cpu time is 0.39 hours to do a 2ps simulation but
> the job ends after almost 3hours even if there are no other jobs running
> on the machine. It seems that each processors are just using only a few
> percentage of the cpu available.
> When I am running the same simulation on only one processor I got 2.26
> hours of cpu, and I really get the result after 2 hours 25.
> this time the program use all the cpu available all the time.
>
> I would like to know if someone has already encounter this problem and
> how I can fix it ?
>
> Sincerely, 

--
==========================================================================
Alexandre Gillet

graduate student
Mathematique Informatique et Genome,
Batiment des Biotechnologies,           Email: gillet_at_glycine.jouy.inra.fr
INRA, Jouy-en-Josas, 78352, France