Mpiexec release 0.76, MPICH2 support, -nolocal, more

[Christopher D. Maestas]

Hello,

Are the mpiexec patches in the torque release now?

[cdmaest at capdesk torque-1.0.1p6]$ patch -p1 --dry-run <
/tmp/mpiexec-compile/mpiexec-0.76/patch/pbs-2.3.12-mpiexec.diff
patching file README.mpiexec
patching file src/cmds/pbs_demux.c
patching file src/lib/Libifl/tm.c
Hunk #1 succeeded at 458 (offset 7 lines).
patching file src/resmom/start_exec.c
Hunk #1 succeeded at 1858 with fuzz 1 (offset 469 lines).
Hunk #2 FAILED at 1912.
Hunk #3 FAILED at 2136.
2 out of 3 hunks FAILED -- saving rejects to file
src/resmom/start_exec.c.rej

It looks like there's some issues with the last hunk of start_exec.c
patching ....

Regards,

On Mon, 2004-04-19 at 13:02, Pete Wyckoff wrote:
> There are not too many changes from 0.75 since it has only been
> four months, but there are a few new features and a fix or two.
> 
> New command-line option "-nolocal"
> 
>     This option is used to avoid running a process on the local machine,
>     that is, the one running the PBS batch script.  Patch with
>     documentation provided by Ron Brightwell.  Note that this option is
>     not available for MPICH/P4 due to the way the execer interface in
>     the library always connects to localhost.
> 
> New communication library: MPICH2/PMI
> 
>     This launch interface promises to be the only one used by the MPICH2
>     library but there may be changes while MPICH2 is still in beta
>     testing.  Version 0.96p2 on x86 with sock:ssm seems to work,
>     although you get two annoying debug messages from each process at
>     startup.  (Note that the seven other codes called mpiexec in the
>     mpich2 distribution are not related to this one.)
> 
> Fix nodect and ncpus
> 
>     Handle more gracefully the case in which PBS provides both "nodect"
>     and "ncpus" entries in the job Resource_List.  You can get yourself
>     into this situation with the PBSPro version or by explicitly setting
>     both those flags with -l, not that you would want to do so.
> 
> Aggregate "task died" warnings
> 
>     The multiple warnings are now combined when all processes exit
>     similarly, whether with an exit code or a signal.  The long listing
>     of one line for each process was becoming onerous for highly
>     parallel run failures. 
> 
> Switch to autoconf 2.57
> 
> Minor fixes for SGI MIPS compiler.
> 
> 
> Full changelog and downloads at:  http://www.osc.edu/~pw/mpiexec/
> Send bug reports, comments and suggestions to the mailing list.
> 
> 		-- Pete
> 
> _______________________________________________
> mpiexec mailing list
> mpiexec at osc.edu
> http://email.osc.edu/mailman/listinfo/mpiexec