RE: notes on upgrading to 2.1.0p0

Hi Garrick:

 

Does that mean LAM/MPI and Open MPI will also be affected?

 

Thanks,

 

Bernard

---

From: torqueusers-bounces@xxxxxxxxxxxxxxxx on behalf of garrick@xxxxxxxxxxxxxxx
Sent: Thu 11/05/2006 17:04
To: torqueusers@xxxxxxxxxxxxxxxx
Subject: [torqueusers] notes on upgrading to 2.1.0p0

Just some quick notes on upgrading to 2.1.0p0:

  - disk and wire formats are unchanged from 1.2.0 or 2.0.0, so feel
    free to upgrade in bits and pieces.

  - 2.1.0 has an entirely new build system supporting shared libs,
    new configure options and defaults, and writes a 'pbs-config' to
    help 3rd party programs build.  Be sure to check './configure --help'
    before building.

  - 3rd party programs (maui, perl-PBS, mpiexec, etc.) need to be
    updated to build against the new shared libs.  mpiexec-0.81
    perl-PBS-0.29 will work.  maui and pbs_python should be coming
    soon.  I've sent patches to other developers but haven't heard
    back yet.

  - existing 3rd party programs statically linked to 2.0.0's libs
    should generally work fine.

  - This is my "live cluster" upgrade procedure:
    1. install new packages everywhere
    2. kill your scheduler
    3. wait for newly started jobs to finish starting (should have
       non-zero walltime)
    4. restart pbs_server
       - check torque.cfg, and set the following new server attributes
         as required:
         server_name, allow_node_submit, submit_hosts
    5. make sure nodes are happy and new jobs can be started with qrun
    6. restart pbs_mom on idle nodes, make sure they are happy
    7. mark busy nodes offline (pbsnodes -o nodename)
    8. start the scheduler, make sure everything is happy
    9. restart offline MOMs as jobs exit and clear their offline bit
       (pbsnodes -c nodename)

  - 2.1.0 has a new "automatic MOM restart" feature that is disabled
    by default.  See 'enablemomrestart' in the pbs_mom manpage.

_______________________________________________
torqueusers mailing list
torqueusers@xxxxxxxxxxxxxxxx
http://www.supercluster.org/mailman/listinfo/torqueusers

_______________________________________________
torqueusers mailing list
torqueusers@xxxxxxxxxxxxxxxx
http://www.supercluster.org/mailman/listinfo/torqueusers