[Dunder-users] dunder: rolling upgrade

[Per Lundqvist]

The compute nodes on Dunder will be upgraded during this week, beginning 
from today. This will be a rolling upgrade, without stopping the batch 
queuing system or any other vital services on dunder (similar to previous 
2 upgrades).

The following will be upgraded on n1-n50:

1. New kernel, 2.6.9-55.0.9.EL_lustre.1.6.3smp
2. Security updates to other software (latest CentOS 4)
3. New lustre client: 1.4.11.1 -> 1.6.3 
4. Upgrade of infiniband stack OFED 1.1 -> OFED 1.2.5.1
5. Upgrade of Scali MPI 

This has been thoroughly tested without any issues, but contact 
smhi-support at nsc.liu.se if you encounter anything after the upgrade that 
does not work properly.

Update Procedure:

1. _All_ compute nodes will be marked unavailable in the batch queuing 
   system (prevents new jobs from starting on the nodes and does not 
   affect already running jobs)
2. For all unavailable compute nodes, and if no job is running on node, 
   and not updated: update 
3. Mark node available again when update finished

The login node and the system node will be updated later.

/Per

-- 
Per Lundqvist

National Supercomputer Centre
Linköping University, Sweden

http://www.nsc.liu.se