What is difference between Hydra and Torque and what is better: MPICH2 or OpenMPI

Question

I have two questions:

1. What is difference between Hydra and Torque, or to ask in other way: what more Hydra have to offer in compare to Torque? Do I need Hydra at all if I choose to use Torque (+ MAUI)?

2. Also, what is an advantage of MPICH2 in advance of OpenMPI, since OpenMPI is supporting IB and also have continuously supporting Windows platform? For me it looks like swiss knife. Am I wrong?

Answer 1

Torque and Hydra are two completely separate things. Torque is a distributed resource manager that allows execution in batch mode of tasks (jobs) on a network of compute systems. Hydra is part of MPICH and is responsible for launching and controlling processes that are part of the MPI job. The way Torque and Hydra work together is that one submits a job to Torque, which reserves cluster resources and at some point start the job. The mpiexec command in turn uses Hydra to start and control the processes that make the MPI job on the compute nodes provided by Torque.

MPICH2 and Open MPI are both quite mature MPI implementations. While Open MPI supports more connection protocols, there is an InfiniBand-enabled version of MPICH called MVAPICH. MPICH is also basis of several commercial MPI implementations including Intel MPI and Microsoft MPI. While Open MPI has supported Windows for a long time, their Windows maintainer left some time ago and it is unclear if they will continue to support that OS.