From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Thu Feb 05 2015 - 10:32:28 CST
If you have "margin" set to something, remove it, since the only role of
the margin parameter is to make the patches larger (and hence the number
of patches smaller.
You can use "twoAwayX yes" to double the patch grid in one dimension,
which is done automatically at some processor count based on heuristics.
Beyond that, if you are running on a cluster see if you can reduce your
node count slightly without losing performance based on the suggestions at
On Thu, 5 Feb 2015, Ryan Gordon wrote:
> I was wondering if there was a way to specify what you want the patch grid dimensions to be? I am running a NPT system of roughly 110,000 atoms, and I allow this to run at a temperature close to the solvent boiling point for 100 ns. I then quenched the system at three different time points from this simulation to room temperature. The way I did this was to decrease the temperature by 10K for 100 ps (enough time for the volume to equilibrate), until I reached my desired temperature (293K). When I run at 300K and up the box size is big enough to run on a patch grid of 8 by 8 by 8. When I reduce to 293K, the box size shrinks just enough to run on 7 by 7 by 7 patch grid, which causes a performance hit since there are more atoms per patch now. My first theory is to make the system slightly larger to keep the system in the 8 by 8 by 8 patch grid dimension, to optimize the computational time to complete this. This seemed tedious if there was a way to manually adjust the patch grid dimensions.
> Thank you,
> Ryan GordonPh.D. Candidate
> Chemical and Biological Engineering
> Drexel University
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:20:53 CST