HPC2N
High Performance Computing Center North
Theano 0.9.0 with full GPU and cuDNN support is now installed on Kebnekaise.
See http://deeplearning.net/software/theano/NEWS.html for what's new.
srun works again for parallel jobs on both Abisko and Kebnekaise, so in general you should no longer use mpirun.
Note that srun and mpirun are now equivalent in most cases, but mpirun does not always work when using standard input (mpirun program < file).
During 2017-06-07 - 09 we will perform a major upgrade of the disk subsystem serving the /pfs/nobackup file system.
To do this, the file system must be quiet and no jobs can run, nor can anyone access the file system from the login nodes.
Jobs will not be allowed to start unless they can finish before the maintenance window starts at 08:00 on the 7th.
The final upgrade of Abisko to Ubuntu Xenial will take place on May 23rd.
We will then upgrade the batch server system.
This is expected to take most of the day if all goes well. If there is trouble it will take longer.
During this upgrade there can be no jobs running on the system. We have therefor put a system wide reservation in place starting 2017-05-23 08:00.
Any jobs that can't finish before that time will not be allowed to start.
The Intel Xeon Phi 7250 (KNL) nodes are finally available for general usage. Everyone with a project on Kebnekaise have access to the nodes and should be able to compile, build and submit jobs to the KNL-nodes.
For instructions how to use the KNL-nodes see Using the Intel Xeon Phi 7250 (KNL) nodes.
A more general description of the KNL architecture is a available from our Overview of the KNL architecture.
The temporary abisko-x.hpc2n.umu.se address has been removed.
Please use abisko.hpc2n.umu.se instead
The ordinary login node of Abisko is now online again running Xenial with a similar setup as the one used on Kebnekaise.
There is one other user visible change and that is that the Abisko login node has gotten a new ssh host key.
Most of the ssh clients will complain about this if they have saved the old key.
A message looking something like the following will be seen (this is from a linux ssh session and might be different if using putty or wome other tool):
Abisko is now back in production after the extended maintenance.
Unfortunately a unexpected problem with the upgrade of SLURM caused all pending jobs to be discarded.
We are therefor allowing jobs to be submitted to the old (Precise) part of Abisko for one more week (until April 28th). After that only jobs submitted from the Xenial (abisko-x.hpc2n.umu.se) node will be accepted into the queue.
The system has also been "split" in two halves, one half running Precise the old way, and one half running Xenial and thus only usable from abisko-x.hpc2n.umu.se.
Kebnekaise is now finally back in production after the somewhat delayed maintenance.
Main changes are an updated SLURM version and a new Lustre file system driver version.
On Wednesday April 19th we have a maintenance window from 08:00 to 21:00 to change some Lustre configuration and update Slurm on Abisko.
This affects the /pfs/nobackup filesystem on all batch and login nodes on both Kebnekaise and Abisko.
We must therefore drain the batch nodes of jobs. This means that there will be gaps which can be filled by short jobs during the days before the 19th.
Please do submit short jobs to fill those gaps.
