mpiexec - configuration

[Pete Wyckoff]

javier.crespo at itp.es said:
> I have a PC cluster of dual processors Pentium IV  and built-in Red Hat 7.3 (kernel 2.4.18-4smp).
> This cluster's jobs queue is managed by a PBS 2.3.16 that works correctly without running mpiexec;
> this week I  have just configured 'mpiexec' with the options
>        --with-pbs=/usr/local/pbs/lib
>        --with-smp-size=2
>        --with-default-comm=mpich-p4
>        --with-mpicc=/usr/local/mpich-1.2.4/bin/mpicc

Looks fine (but the --with-pbs probably is unnecessary).

Did you patch PBS?  Not necessary, but you may want to eventually.

Which version of:  mpich, mpiexec?

Only mpich-1.2.4 and later work with mpiexec using p4.  You might want
to grab the latest mpiexec-0.70 too.

> and the testing program shows a heap of "unspected lines" as the followings
> 
>            process not in process table; my_unix_id = 32023 my_host=tecws003
>            Probable cause:  local slave on uniprocessor without shared memory
>            Probable fix:  ensure only one process on tecws003
>            (on master process this means 'local 0' in the procgroup file)
>            You can also remake p4 with SYSV_IPC set in the OPTIONS file
>            Alternate cause:  Using localhost as a machine name in the progroup
>            The names used should match the external network names.
>            p4_error: p4_get_my_id_from_proc: 0
>            p4_error: rm_start: net_conn_to_listener failed: 40839
>            mpiexec: Warning: main: task 1 exited with status 1 (raw 0x1)

This is all P4 complaints.  Perhaps you did not compile mpich to enable
shared memory with P4, and mpiexec is starting multiple copies per node
causing the P4 library to complain?

> When trying to run a parallel program of my own (that goes well with mpirun) the output error file
> contains
> 
>            mpiexec: Error: get_hosts: pbs_connect: No server specified.

This usually means that /var/spool/pbs/server_name does not exist,
although you should check in whatever the pbs spool directory is at your
site.

There may be lots of issues going on here.  Hope the above helps you
figure it out a bit more.

		-- Pete