Abisko - down for transistion to the final system

The time to transition Abisko from the interim to the final system is here. During the
transition, Abisko will be unavailable starting from Wednesday February 15th 08:00. Assuming the transition goes smoothly, we expect to have the final system functional early next week.

During the migration from the interim system, we need to make a few changes that are disruptive to the cluster interconnect. The batch nodes will therefore not process any jobs from the above point in time until the transition is complete.

The architecture of the final system is slightly different from the interim system.
Consequently there is a risk that some jobs compiled on the interim system will not run
properly (e.g. fail), including jobs currently sitting in the queue. User compiled code
should be recompiled anytime after the login node comes back online to ensure compatibility with the new architecture. Jobs not depending on user compiled code and which only depend on HPC2N installed binaries should function properly.

The Abisko login node (abisko.hpc2n.umu.se) will be down from 8:00 for a short while. The node will come back up later that day with the new architecture, ready for recompiling.

The final system will consist of 318 nodes, each node having 4 x AMD Opteron 6238 (Interlagos 2.6 GHz, 12 core), resulting in a total of 15264 cores. Of those nodes, 10 are fat (512 GB RAM/node), and 308 thin (128 GB RAM/node). Thus the final system will have 48 cores per node, whereas the interim system only had 32 cores per node. The peak performance of the system is calculated to 158.75 TFlops.

More information about Abisko can be found here: /resources/hardware/abisko.

We apologize for the delay in bringing the final system online.

If you have any questions please visit our support website, or contact us via support@hpc2n.umu.se.

Updates will be posted here as the transistion progresses.

Wed, 2012-02-15 12:19 | Birgitte Brydsö