[gmx-users] reasons for slow computation?
Justin A. Lemkul
jalemkul at vt.edu
Mon Oct 10 21:34:31 CEST 2011
Yun Shi wrote:
> Hi all,
> I am doing MD simulation on two almost identical protein-ligand systems
> with GROMACS4.5.4 and amber99SB force fields.
> Almost every single parameter I used for this two systems are the same
> (I literally copied the mdp files for both), except that one has 65235
> atoms while the other has 65205 atoms, which is caused by the difference
> of 10 water molecules.
> However, the 65235-atom one is significantly slower than the 65205-atom
> one, the computational speed (performance) ratio is roughly 1:3.
> I am really confused, since the only differences between them that I can
> think of are:
> 1. The configuration of ligands, that is, alpha sugar vs beta sugar
> 2. number of water molecules, 19477 vs 19467
> 3. octahedral box size, ~726.119 nm^3 vs ~726.322 nm^3
> 4. probably the initial velocities, since I am used gen_seed = -1
> 5. the cpu specs, as I am using a cluster (I specified procs=72 for both
> system and I assume they were assigned to different cpus randomly),
> which consists of Intel Xeon E5450 quad-core processor, running at 3.0
> GHz, and Intel Xeon X5650 6-core processors running at 2.67 GHz
Point #5 is the one to check. You *assume* they are running on different
processors, but are they? Are the nodes that are running the slow job fully
functioning, or is one (or maybe more) down or overburdened with some other
process(es)? Try running the jobs independently so you can be sure there is no
potential overlap. There is no Gromacs-based reason for this to be happening.
I strongly suspect the hardware is the problem.
> How come the performance could have a 1:3 difference?
> Thanks for any suggestion!
Justin A. Lemkul
ICTAS Doctoral Scholar
Department of Biochemistry
jalemkul[at]vt.edu | (540) 231-9080
More information about the gromacs.org_gmx-users