WIP: Migrate Baobab to CentOS7

HPC Announce HPC ChangeLog
1

Since a long time ago, we plan to upgrade Baobab to CentOS7. The principal issues with the current version of the OS are:

  • GLIBC version 2.12
  • Kernel 2.6.32-754.3.5.el6.x86_64 (still maintained/patched but quite old)

We weren’t able to do this because it’s a huge step to go from CentOS6.10 (the latest version) to CentOS7 and we didn’t had a spare infrastructure to test the migration. For this reason, we have installed a new login node that will be used for the migration.

The steps are the following:

  1. Install the new login node in an automated way (kickstart, pxe) with CentOS7
  2. Put a couple of nodes on drain, install them in an automated way in CentOS7
  3. Do some testings (SLURM, mpi, cuda, storage)
  4. Ask beta testers that are willing to try the new OS version and login node to use the new login node.
  5. Progressively migrate nodes to the new OS

We will update this post to let you know how it is going.

Task 1 is done (03.03.2019)
Task 2 is done (05.03.2019)
Task 3 is done (06.03.2019)

3 Likes
2

Thanks great new.

I would personally be happy to test CentOS7 gpu node with singularity.

3

Hi there,

the first nodes in CentOS7 are now available (with specific partitions, when applicable):

  • login2 (AKA baobab2.hpc.unige.ch )
    We were obliged to replace GNOME3 with XFCE since the former requires hardware 3D acceleration not available via X2go (upstream bug)
  • node003 and node004 in partition debug-EL7
  • gpu008 in partition shared-gpu-EL7
    Many thanks to the Kalousis group to have accepted to migrate one of their GPU nodes!

For those wanting to test CentOS7, please login to baobab2.hpc.unige.ch and start your job on the -EL7 partitions as usual.

However, please pay attention that while we verified as much as possible the CentOS7 installation, there could still be minor glitches.

Thx, bye,
Luca

4

Hi there,

the first nodes in CentOS7 are now available (with specific partitions, when applicable):

4 more:

  • node[226-229] in partition mono-shared-EL7 and shared-EL7
    Many thanks to the Golling group to have accepted to install their new nodes directly in CentOS7!

Thx, bye,
Luca

5

Hi there,

the first nodes in CentOS7 are now available (with specific partitions, when applicable):

one more:

  • node203 in partition shared-bigmem-EL7
    Many thanks to the Gonzalez-Gaitan group to have accepted to migrate one of their CPU nodes!

Thx, bye,
Luca

6

Current issues and workaround due to the migration to CentOS 7

  • Stata on Baobab isn’t working anymore. You need to use the new module:
    module load Stata/14
7

Hi there,

6 more:

  • node[184-185,196-199] in partitions mono-shared-EL7 and shared-EL7

Thx, bye,
Luca

9

Hi there,

2 more:

  • gpu[009-010] in partitions cui-gpu-EL7

Thx, bye,
Luca

10

Hi there,

5 more:

  • node[063-066,085] in partitions mono-shared-EL7 and shared-EL7

Thx, bye,
Luca

11

Hi there,

14 more:

  • node[082-084,086,128-129,131,140-141,144-147,152] in partitions mono-shared-EL7 and shared-EL7

Thx, bye,
Luca

12

Hello,

58 more !

  • node[077,087-101,103-127,132-139,142-143,148-151,153,159,162] in partitions mono-shared-EL7 and shared-EL7

Sincerly,

Massimo

13

Hi there,

the migration ended on 2019-08-09 (cf. Baobab migration from CentOS6 to CentOS7 ).

Thx, bye,
Luca