Is the PVM library (parallel virtual machine) widely used in HPC?

Did everyone migrate to MPI (messaging interface) or is PVM still widely used in supercomputers and HPC?

+4
source share
2 answers

My experience is that PVM is not widely used in high performance computing. MPI seems to be widely used, and something like a co-array Fortran could be a sequel to the massively parallel systems of the future.

I use a library called InterComm to combine physics models together as separate executables. InterComm currently uses PVM to communicate between these related models. PVM and InterComm boast that they work in homogeneous and heterogeneous network environments (I was told that MPI does not support heterogeneous computing / network environments). However, this is a function that we have never used (and I very much doubt that we will ever be).

I had a hard time working PVM in academic computing environments. Some people from the system administrator / support at reputable national computing centers even suggested that we β€œjust” recode our 20-year-old O (10 ^ 4) code to use MPI because of the problems we encountered when porting code for a particular supercomputer in which the / queing router environment did not like to run multiple parallel executables with PVM.

If you are at the stage of project architecture / design, I would recommend not to move away from PVM if you do not need to work in heterogeneous computing / network environments!

+8
source

It can be highly site dependent, but in my experience MPI completely dominates PVM in the (at least academic) HPC space. You cannot realistically launch a new HPC interconnect without MPI support, but obviously PVM is optional. Is there a PVM implementation for eg Infiniband?

+6
source

All Articles