[gmx-users] HPC performance in Gromacs ?

[Li, J.]

Hello all,

I have a report as following, but I do not have any clear clue about what I
have to do. If I change some default values for -dd/-dds/..., what I should
expect and what I should be careful about about tunning these parameters?

It said the load imbalance is from inhomogeneous. But I am not very clear
what inhomogeneous mean here. My system has two kinds of polymer chains and
spatially no vacuum  gap. Why did not these default values suit for my
system?

And I have chosen yes for -dlb option. If not, all the cut-off distance and
PME grid changed for the load balance. As a consequence, the simulation has
been changed, right?

************************************************************************************************

P P - P M E L O A D B A L A N C I N G

NOTE: The PP/PME load balancing was limited by the domain decompostion,
you might not have reached a good load balance.
Try different mdrun -dd settings or lower the -dds value.

PP/PME load balancing changed the cut-off and PME settings:
particle-particle PME
rcoulomb rlist grid spacing 1/beta
initial 1.300 nm 1.300 nm 144 144 144 0.139 nm 0.416 nm
final 1.300 nm 1.300 nm 144 144 144 0.139 nm 0.416 nm
cost-ratio 1.00 1.00
(note that these numbers concern only part of the total PP and PME load)


D O M A I N D E C O M P O S I T I O N S T A T I S T I C S

av. #atoms communicated per step for force: 2 x 38901.0

Average load imbalance: 57.1 %
Part of the total run time spent waiting due to load imbalance: 3.2 %
Steps where the load balancing was limited by -rdd, -rcon and/or -dds: X 2
% Y 0 %
Average PME mesh/force load: 9.597
Part of the total run time spent waiting due to PP/PME imbalance: 58.9 %

NOTE: 58.9 % performance was lost because the PME ranks
had more work to do than the PP ranks.
You might want to increase the number of PME ranks
or increase the cut-off and the grid spacing.
-- 
Jing Li
PhD student
FMNS, University of Groningen
Room 5115.0309, Nijenborg 4, Groningen
The Netherlands