My questions relate to MPD. The only info I could find in the Intel installations docs were:
"The contents of the machines.LINUX file can also be
used by you to construct an
mpd.hosts file for the multi-purpose daemon (MPD)
protocol. The MPD protocol is used for running MPI applications that utilize Intel
I got the daemons to work and can run MPI programs, but as a system admin I need more info on these. How do they function? Do they need to be run for every user separately? It seems that you can create more than 1 daemon per node... why would you do this??
Is there anywhere I can find more info on MPD and its inner workings?
MPDs are used to create a ring of processes which can control mpi application and communicate with each other to provide needed info. For example MPD processes are used for pinning.
Each user can create own ring and its mean that new mpd shall be started on each needed node. One user can start an application on 2 nodes another user can start an application on 32 nodes...
Usually you just need to allow to a user to start mps processes.
MPDs can be started in 2 different ways: by mpdboot and by mpdrun. In case of mpdboot mpd ring will exist after application is complete, but in case of mpdrun mpd ring will be closed when application is complete.