[gmx-users] Simulating interfaces and load imbalance

[Gabriele Lanaro]

Hi! Recently I've been simulating a system comprised of water and LiCl ions
in different arrangements. Depending on the arrangements I get different
responses in terms of performances. The box is tetragonal (but I heard from
other people in my lab that they had a performance hit also in cubic boxes,
not sure if it's the same as my case anyway).

This is running on 64 processors and I have 51840 atoms. The Li/Cl/H2O are
in equal proportions (10368 each).

Test1: Water and ions well mixed.
Box dimension 30.23817   5.54007   4.84122

A set of DD entries from the log files:

DD  step 13509999  vol min/aver 0.714  load imb.: force  4.3%  pme
mesh/force 3.802
DD  step 13519999  vol min/aver 0.794  load imb.: force  4.0%  pme
mesh/force 1.701
DD  step 13529999  vol min/aver 0.833  load imb.: force  6.6%  pme
mesh/force 1.652
DD  step 13539999  vol min/aver 0.701  load imb.: force 11.7%  pme
mesh/force 2.062
DD  step 13549999  vol min/aver 0.686  load imb.: force  3.8%  pme
mesh/force 1.934
DD  step 13559999  vol min/aver 0.768  load imb.: force  1.9%  pme
mesh/force 1.639
DD  step 13569999  vol min/aver 0.919  load imb.: force  2.1%  pme
mesh/force 1.369

Test2: All ions on one side (a cubic crystal) all the water on the other
(they are divided in the x direction).
Box dimensions 23.03340   6.15600   6.15600

DD  load balancing is limited by minimum cell size in dimension X
DD  step 7569999  vol min/aver 0.868! load imb.: force 39.9%  pme
mesh/force 1.036
DD  load balancing is limited by minimum cell size in dimension X
DD  step 7579999  vol min/aver 0.869! load imb.: force 43.1%  pme
mesh/force 1.030
DD  load balancing is limited by minimum cell size in dimension X
DD  step 7589999  vol min/aver 0.868! load imb.: force 41.8%  pme
mesh/force 1.035
DD  load balancing is limited by minimum cell size in dimension X
DD  step 7599999  vol min/aver 0.869! load imb.: force 41.1%  pme
mesh/force 1.044
DD  load balancing is limited by minimum cell size in dimension X
DD  step 7609999  vol min/aver 0.868! load imb.: force 40.3%  pme
mesh/force 1.030
DD  load balancing is limited by minimum cell size in dimension X
DD  step 7619999  vol min/aver 0.869! load imb.: force 43.8%  pme
mesh/force 1.031

Basically the Test2 is very slow and I would like to find a way to
interpret those messages and try fix this issue (tell me if there's
something I should read).

Thank you,

Gabriele