We have been restarting a set of jobs and the results of the continuation appear fine and perfectly in line with the starting jobs, but have just noticed on going to some longer run times (more time steps) that the restarted jobs run at a pace about 75% slower than the original. The jobs are for a coarse-grained polymer using the Langevin thermostat, and also have solid boundaries (if all that makes any difference). In the restart the all variables, all fixes, all regions, and the velocity command are reissued.
Is there a possible reason/remedy for this?
Fred
We have been restarting a set of jobs and the results of the continuation
appear fine and perfectly in line with the starting jobs, but have just
noticed on going to some longer run times (more time steps) that the
restarted jobs run at a pace about 75% slower than the original. The jobs
are for a coarse-grained polymer using the Langevin thermostat, and also
have solid boundaries (if all that makes any difference). In the restart the
all variables, all fixes, all regions, and the velocity command are
reissued.
can you provide some input that would allow somebody
to try reproduce it?
my first hypothesis would be that this would be related
to a different domain decomposition or rather different
processor grid. but this is something, where i'd like to
see it for myself and also see the exact input and try
out some modifications in order to narrow down potential
other explanations.
Is there a possible reason/remedy for this?
before suggesting a remedy, the reason needs to be
identified. it can be as simple as a runaway process
on (one of) the node(s) you are running on.
axel.