[gmx-users] GROMACS mpirun with extremely high load imbalance in output (vol 0.43! imb F 15- 53%)

[Mark Abraham]

Hi,

It would be easier to offer an opinion if we knew what was in the
simulation system (does load imbalance even make sense?) and what the
performance characteristics are without PLUMED. Because of the way PLUMED
works, we can make no promises as to performance or correctness...

Mark

On Thu, Mar 1, 2018 at 6:58 PM Marc Hoemberger <hoemberg at brandeis.edu>
wrote:

> Hi gromacs-users,
>
> I am currently running a set of TMD simulations with gromacs 5.1.4 +
> plumed_2.4.0. The system for all four runs is identical, the only
> difference is the force constant for bias applied by plumed.
>
> I am using gromacs with mpi support, starting it with 32 processes on my
> 32core processors. While two of the four runs take a relatively short time,
> the other two are estimated to be done 5 days later. The only difference I
> can make out in the gromacs output is that for runs that are running at
> normal speed the output of gromacs reports low imbalance, i.e. fluctuating
> between 0 and 2% :
>
> vol 0.39! imb F  1% pme/F 0.20 step 224700, will finish Wed Mar
>
> However, in the two simulations that are running much slower the imbalance
> reported is much higher, i.e. fluctuating between 10 and 50+%:
>
> vol 0.39! imb F 15% pme/F 0.81 step 4222700, will finish Wed Mar
>
> I left -dlb option of mdrun to auto. Could this difference indeed explain
> the vast difference in calculation time of the different runs, and if so is
> there anything I could do to alleviate this?
>
> Best,
> Marc Hoemberger
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>