Link to home
Create AccountLog in
Avatar of jlewi
jlewi

asked on

Starting multiple MPI jobs using MPD

I'm using mpd to manage MPI jobs on a cluster.

If I want to run an MPI job I run mpdboot to start the mpd daemon on all the nodes.

If I want to start multiple MPI jobs I run mpdboot once for each job. Will running mpdboot multiple times cause multiple mpd daemons to be started? If multiple mpd daemons are started is that a problem?

When I finish a job, I run mpdallexit to clean up the mpd daemons started. Will this kill all mpd daemons including the ones started for other jobs?

So in summary, if you want to run multiple MPI jobs on the same machine what is the proper way to start and stop the MPD daemons?
Avatar of Duncan Roe
Duncan Roe
Flag of Australia image

If MPI is anything like LAM (the only cluster software I'm familiar with) then the answer is: start all the daemons at boot time and never clean them up.
Starting them again may start new ones or may not - preferably not. Stopping them is very bad practice - you might well shut down the only one and your other running cluster applications could then fail.
ASKER CERTIFIED SOLUTION
Avatar of jlewi
jlewi

Link to home
membership
Create a free account to see this answer
Signing up is free and takes 30 seconds. No credit card required.
See answer