[gmx-users] equilibrium stage runs for too long
jalemkul at vt.edu
Thu Aug 23 19:33:03 CEST 2012
On 8/23/12 1:20 PM, Clare wrote:
> Dear all,
> I'm performing NVT on a system consisting a water layer, a decane layer and a
> lipid layer in between those two layers. There are 40 lipids, 627 decane
> molecules and 8577 water molecules in total.
> The EM was conducted successfully, however, when I try to run NVT on this
> system, it seems to take forever. I'm using 48 cores on a super computer and the
> report suggests it has token 367 hours already for the performance of one core,
> and it still has not finished yet.
Depending on the size of the system (number of atoms), you may not be getting
good scaling. You should aim for between a few hundred and a few thousand atoms
per core. The vast amount of time taken suggests something else more
fundamental is wrong. A rough guess would suggest this system should take no
more than a few hours (less than 5) on any modern hardware with reasonable scaling.
Being that we don't know how you're invoking the command or how your hardware is
configured (consult your sysadmins, not us) it's hard to say more.
> So I'm wondering does it really require so much time to run or is there
> something wrong in my system which prevents it to converge? Additionally,
> there's a running time specified in the mdp file which in my case is 100 ps, so
> I suppose the simulation should end when the time limit is reached regardless of
> whether the pressure converges to the reference value?
Nothing converges automatically during an MD run. It is not like EM that will
stop when a tolerance is reached. mdrun will perform the prescribed number of
steps. Whether or not that is sufficient to equilibrate the observable(s) of
interest is up to you to assess after the run.
Justin A. Lemkul, Ph.D.
Department of Biochemistry
jalemkul[at]vt.edu | (540) 231-9080
More information about the gromacs.org_gmx-users