openSUSE 13.2 Infiniband connection?

Dear openSUSE users,

I am running a few identical servers with openSUSE 13.2 for some quantum chemistry computation and I would like to combine these servers together in a small cluster.

The networking hardware that I already have:
(1) Mellanox ConnectX-2 VPI Dual Port HCA Infiniband QDR adapter on each server,
(2) Mellanox 4036E 36 Port QDR switch,
(3) 4 identical servers with same CPUs, memory and GPUs as computing nodes,
(4) QDR copper cables.

I read Mellanox driver webpage, http://www.mellanox.com/page/osv_support_ib#InfiniBand

But it seems openSUSE 13.2 is not listed and I don’t want to reinstall the OS to SLES 12 SP2.

Does anyone try installing MLNX_OFED on openSUSE? Does it work?

Thank you,
Rolly

Educated guess here…based on the support matrix page for MLNX_OFED 3.4-1.0.0.0

RHEL: 6.2,6.3, 6.5, 6.6, 6.7, 6.8, 7.0, 7.1, 7.2
CentOS: 6.2,6.3, 6.5, 6.6, 6.7, 6.8, 7.0, 7.1, 7.2
SLES: 10 SP3; 11 SP1, SP2, SP3, SP4 SP4 SAP; 12 12 SP1
OL: 6.5, 6.6, 6.7; 6.8, 7.1, 7.2
Ubuntu: 12.04.5, 14.04, 14.10, 15.04, 15.10, 16.04
Fedora: 19, 20, 21, 22, 23, 24
Debian: 7.6, 8.0, 8.1, 8.2, 8.3

which encompasses kernel versions from 3.2 to 4.8 (based on Ubuntu and Fedora versions mentioned) I would say you’d have no problem with openSUSE support for version 13.2, however I would consider using Leap assuming no other constraints around using the version you’re proposing.

Anecdotal support for Leap 42.1 mentioned here…

Hi Deano,

Thanks. The link is all I found by google on this topic ans I am new to clustering HPC so please allow me to explain my project.

My configuration is here, and I would like to make sure my design is correct.
https://drive.google.com/file/d/0B6geTmCEpzU8SkRKTXI1RHdXMTA/view

My installation steps are,
(1) install the MLNX_OFED on each server and configure the infiniband over IP address, i.e. 192.168.1.1 (4036E), 192.168.1.2 (NFS), 192.168.1.3 to 192.168.1.6 (each compute node)
(2) allow ssh passwordless login from each node to NFS and vice versa.
(3) install intel parallel studio XE on each compute node and write a machine_node file containing each node’s IP address over infiniband
(4) compile the software using intel MPI support

Please check if I miss any step and the above are correct. :slight_smile:

Thanks again,
Rolly

I know very little about InfiniBand. I was only answering about the likely kernel support.

For driver compilation, you’ll likely need the following packages installed at least

zypper install gcc make patch gcc-c++ kernel-syms

This might be helpful as far as an overview
https://wiki.archlinux.org/index.php/InfiniBand#Hardware

Good luck