tm_init: tm: not connected, protocol version 2?
Pete Wyckoff
pw at osc.edu
Thu May 29 20:29:38 EDT 2003
lnthurston at ucdavis.edu said on Thu, 29 May 2003 15:41 -0700:
> I have installed mpiexec 0.74 on a cluster that is running OpenPBS
> 4.3.16. I was using PBSPro and was able to get mpiexec to run, but was
> not happy with the inability to feed standard input to the parallel
> processes. I therefore decided to try OpenPBS. Unfortunately I have
> been unable to get mpiexec to run under OpenPBS.
>
> In the mom_log on the node there is the following error...
>
> 05/29/2003 11:18:49;0001; pbs_mom;Svr;pbs_mom;Success (0) in
> tm_request, bad protocol version 2
A crucial bit of information there. This is one of the pbs_mom
processes on a compute node saying that something tried to talk to it
using the wrong TM version. My openpbs 2.3.16-ish source tree says that
the version should be 1, not 2, hence it appears that you may be running
the OpenPBS mom.
But, could you have linked mpiexec against -lpbs from your PBSPro
distribution instead of from your OpenPBS version? If you don't tell
configure specifically, it tries to use /usr/local/pbs/lib/libpbs.a.
Add "--with-pbs=/where/ever" to the configure invocation line to change
that.
-- Pete
More information about the mpiexec
mailing list