High performance cluster Ukko

Computer Science Department's high performance cluster, Ukko, is available to all department's users. You can check the list of nodes and their status and load from http://www.cs.helsinki.fi/ukko/hpc-report.txt. Public nodes are available using CS user accounts and work exactly like CS department's interactive Linux servers.

The nodes are named from ukko001.hpc.cs.helsinki.fi to ukko240.hpc.cs.helsinki.fi. They can be accessed from inside university's network and from university's openVPN service with ssh.

Sometimes the nodes need to be rebooted for software or firmware updates. Reboots are scheduled to happen during workdays, between 12:00 and 16:00, one node at a time and with five minute intervals, so that only few nodes should be unavailable at any given time (however sometimes a reboot takes longer than 5 minutes).

  • Ukko nodes from ukko001 to ukko048 are rebooted on Mondays
  • Ukko nodes from ukko048 to ukko096 are rebooted on Tuesdays
  • Ukko nodes from ukko097 to ukko144 are rebooted on Wednesdays
  • Ukko nodes from ukko145 to ukko192 are rebooted on Thursdays
  • Ukko nodes from ukko193 to ukko240 are rebooted on Fridays

The reboot schedule is available from the hpc-report page. If you have a long running job, which would be stopped by a reboot, please contact itsupport@cs.helsinki.fi so that the reboot can be canceled. However, try to avoid starting long running jobs, which would be stopped by a reboot.

Nodes can also be reserved for a specific purpose so that only some users can access it. Please contact department's IT support staff if you need to make a reservation.

Matlab is available in all cluster nodes.

Please note that the cluster is still a work in progress. Virtual host installation, node monitoring and usage policies still need to be properly addressed. If you have suggestions, don't hesitate to contact the IT support group. However, there is no dedicated admin for the cluster, so there might not be enough time to implement all improvements.

The cluster has 240 Dell PowerEdge M610 nodes. All nodes have 32GB of RAM and 2 Intel Xeon E5540 2.53GHz CPU:s. The CPU:s have 4 cores, so one node has a total of 8 cores. The cores are using hyperthreading so there can be a total of 16 threads running at the same time. All nodes have two bonded 10Gbit/s network interfaces.

The cluster has a shared NFS directory available at path /cs/taatto. The old /cs/ukko share is now reserved for legacy users only.  Personal and group subsirectories will be created at /cs/taatto/home and /cs/taato/group on  request and there is also directory /cs/ukko/scratch where everyone has write access. Please remember that old files will be removed from /scratch/ directory to reclaim disk space.  Disk quotas are enabled in the /cs/taatto/ share.

Since 13.3.2013 the /cs/taatto share is also available in interactive host melkinkari.cs.helsinki.fi, which is accessible directly from the internet through ssh (and therefore through sftp and rsync too). Also, it does not have the 100Mbit/s external throughput limit, which is caused by some old routing hardware between Ukko Cluster and CS department.

Please, remember to remove your unused files.

 

Since 1.12.2011 there are also memory usage limits (30 GiB) and process number limits (256) enabled by default, just to prevent accidental crashes of nodes. These are only soft limits: they can be overridden from the command line with the ulimit shell builtin command. Please be careful with your processes and remember that the cluster has other users too!

13.03.2013 - 16:48 Jani Jaakkola
06.10.2010 - 16:34 Jani Jaakkola