[mvapich-discuss] hyrda mpiexec vs. mpirun_rsh for large scale jobs

[Dhabaleswar Panda]

Hi Walid,

Thanks for your note.

The MVAPICH team recommends the use of mpirun_rsh because it delivers best
performance and scalability for job launching on large-scale clusters.
Many large-scale clusters (including TACC Ranger with 64K cores) use it.
Support for Hydra is provided as an alternative. Hydra is designed by the
MPICH2 team at ANL. Please contact them for any details on the limitations
you are experiencing.

Best Regards,

DK


> Dear All,
>
> what is the best mpi launcher for large scale jobs, are there any
> advantages/disadvantages of using mpiexec over mpirun_rsh, the issue is
> that our cluster scheduler is UGE, and they reccomend and advise to use
> hyrda as it provides tight integration out of the box, however we saw that
> it is limiting us to 2600 cores max, we are still investigating the root
> cause, however one finding was that if we use mpirun_rsh there are a couple
> of paramters that allows large scale jobs to run, could not find the same
> for mpiexec.
>
> so, can hydra launch more than 2600 cores?
> should we revert back to mpirun_rsh and solve our tight integration with
> UGE?
>
> regards
>
> Walid
>