Azure HPC images – SLES 12 SP3

Share
Share

Hot on the heels of the SUSE Linux Enterprise Server 12 SP3 release and well before Thanksgiving SLES 12 SP3 based HPC images with RDMA support are now available in Microsoft Azure.

Lets get the upgrade story out of the way first. First get your system up to the latest version of the packages, as in the SLES 12 SP3 blog commands need to be run as root.

zypper up

HPC images have the RDMA driver repos preconfigured and this repo is not located on the region local update infrastructure and therefore not managed by the repository service associated with those servers. Manual intervention is required.

sed -i s/sp1/sp3/ /etc/zypp/repos.d/msft-rdma-pack*.repo

The SLES 12 SP1 HPC images did not have the HPC module product package installed as the HPC module did not exist at the time of release. Thus migration has no special cases, just

zypper in zypper-migration-plugin

zypper migration

2 migrations will be offered you need to choose the migration to SLES 12 SP3. SLES 12 SP2 is broken, which is why there were no SLES 12 SP2 HPC images, more on this later, for those interested.

Accept the de-installation of

msft-rdma-kmp-default-20160329

and the vendor change for all “products:“. Then let zypper go to work. Once zypper is done it is time to reboot.

shutdown -r now

And you probably want he HPC module enabled:

SUSEConnect -p sle-module-hpc/12/x86_64

A word about MPI; As with the SLES 12 SP1 based HPC images the SLES 12 SP3 based image has the Intel MPI runtime libraries available in /opt/intelMPI/intel_mpi_packages/. The libraries are provided in the version 5.0.3 release. Newer versions of the Intel MPI libraries are available from Intel and depending on your application or other needs you will probably pick the version that works for you.

Last but not least, for those interested a few words about why there were no SP2 based HPC images. The answer is actually quite simple, race conditions. Every time we thought we were ready to release, which was about 3 or 4 times during the SP2 cycle we ran into yet another issue. By the time the final fix for the last race condition was in place we were so close to SLES 12 SP3 that a release of a SLES 12 SP2 HPC image simply didn’t make sense. Work is ongoing to avoid this from happening again, with projects on the way to do more and earlier testing as well as more consistent testing to look for the markers that tripped us up in SP2. And with that, we are happy that we are finally able to provide new HPC images and a migration path for running instances in the Azure HPC environment, ENJOY!

Share
(Visited 1 times, 1 visits today)

Leave a Reply

Your email address will not be published. Required fields are marked *

No comments yet

Robert Schweikert
2,083 views