[ofa-general] verb level interoperability between vendor's hcas

Scott A. Friedman friedman at ucla.edu
Fri Jun 26 17:58:24 PDT 2009


We have a largish cluster that has nodes with HCAs from multiple vendors
- specifically, Mellanox and QLogic. The cluster mostly consists of
Mellanox cards and a much smaller number of QLogic cards (all DDR). We
have noticed that there are sometimes problems (crashing) running jobs
between nodes that mix vendors - when using verbs or using libraries
that use verbs (e.g. MPI).

The question I have is actually more general for the recipients of this
list - what kinds and how much interop testing is there? Is this
information available someplace and I am just missing it?

Is a mixed HCA environment cluster not ready for prime time - yet?

Anyone with experiences or comments they would like to share?

Thanks,
Scott



More information about the general mailing list