Author Archives: lsz075

About lsz075

IT-avdelingen

•
For all fimm users :

GPFS file system on fimm crashed today around 10:15, home file system and work file system was unmounted from all compute nodes there for all jobs running were killed.

Problem is resolved now, file system is back online, all crashed jobs has to be resubmitted.

We are apologize for inconvenience.


•
Hexagon will have scheduled maintenance August 1st extending into August 2nd.

We will do changes to all the PDUs to increase reliability of the system and additionally install latest update of the OS (CLE and SMW). The latest software update will also increase stability and decrease startup and debug times for system failures.

The scheduled downtime will start August 1st at 09:00.

Please send any questions to support-uib@notur.no

An email concerning this was sent out on July 25th to all hexagon users.

Hexagon Sysadmins

Updates:

Update Thursday 10:30, The changes to the PDUs is taking longer than expected, therefore the maintenance will be extended into Friday August 3rd.

Update Friday 12:00, There will unfortunately be a further delay before the system is up again. Currently, we expect to boot the system on Saturday August 4th.

Update Sunday 19:25, The maintenance has been finished and the system is back online.

•
Hexagon has updated software/libraries. Please read the full information about new features and bugs fixed in:

http://docs.cray.com/books/S-9401-1206/

In short the following has been updated:

xt-mpich2 (MPI) 5.4.5 -> 5.5.1

PGI 12.4.0 -> 12.5.0
GCC 4.6.3 -> 4.7.0
Cray compiler CCE 8.0.5 -> 8.0.6
xt-asyncpe 5.10 -> 5.11

xt-libsci 11.0.06 -> 11.1
Trilinos 10.8.3.0 -> 10.8.3.1
Petsc 3.2.01 -> 3.2.02
TPSL 1.2.00 -> 1.2.01
fftw 3.3.00 -> 3.3.01
Netcdf 4.1.3 -> 4.2.0
HDF5 1.8.7 -> 1.8.8

Note that Netcdf 4.2.0 no longer provides the legacy libnetcdf_c++ API but only the new libnetcdf_c++4 API.

Programs needs to be recompiled to gain any features or bugfixes due to static linking.

•
Hexagon lost 1 cabinet on May 30th because of a power failure, due to the high resiliency it continues to run. On June 7th at about 08:00 another cabinet also got a power failure. The current state is that it continues to operate but 2 login-nodes are down causing connection attempts to fail (depends on round-robing of dns) and some of the nodes have communication problems. We are investigating possible solutions.

Update 12:30: We need to restart the machine to be able to bring it back up.

Update 13:00: Machine is now up again.