[gmx-users] Re: multi cpu problem
Wei Fu
fuwei at adrik.bchs.uh.edu
Thu Jan 29 06:38:01 CET 2004
Hi, David,
Thanks for your reply. You mean I can do SD with multi processors only in
linux with cshell. Am I right? Anyway, I do a test on sgi workstation, it
turns out: MPI: bad process count.
In addition, I did some extra tests. Here is my summary:
For SD, it can run with 1 cpu, can't run with multi cpu on several
proteins I tested (I tested several proteins).
For MD on only protein, as Santarossa said, it can run with 2 cpu, but
can't run with more than 2 cpus, is it strange?
Cheers,
Fu, Wei
************************************************************
Wei Fu (Linda)
Postdoctoral Research Associate
Department of Biology and Biochemistry
University of Houston
4800 Calhoun Rd. HSC #402C
Houston, TX 77204-5001 USA
Phone:713-743-8355(o) 713-795-9270(h)
Fax:713-743-8351
Email: fuwei at adrik.bchs.uh.edu
http://adrik.bchs.uh.edu/~fuwei/
**************************************************************
> As you stated in the "Subject: Re: [gmx-users] errors depending on
(lack of) solvent", the limitation is in solving (distance) constraints
(e.g., for constant bondlengths). That means that I can run SD with muilt
processors when setting the "constraints = none"?
>
> I have done some SD tests on my protein by "constraints = none", the
results are:
> it runs when submitting job directly (interactive) with 2 cpu
> but it can not run when submitting job to background (by nohup) with 2
cpu.
>
> when set parameter to "constraints = hbonds"
> it can't run SD simulation on my protein with 2 cpu on both interactive
and background mode, but SD can run with 1 cpu.
>
> So, whether this means I only can run SD with 1 cpu? My proteins
consist of 423 residues, 3431 atoms. How long it will take to do a
simulation of 500 ns with 1 cpu?
>
> By the way, whether there is some way to run SD on multiple cpu?
>
> Thanks!!!
>
> Fu, Wei
you have to use mpirun, and put the stuff in the background
mpirun -c 2 mdrun >& logfile &
for LAM and cshell.
You can also use constraints = h-bonds
More information about the gromacs.org_gmx-users
mailing list