Did you ever get this sorted out, if you need any help let me know I have ib working in windows server 2012r2 and 2016 tp5
What about on CentOS7.3 .. a main issue is sorting why 7.2 has IB performance issues in datagram mode (apparently not on Connect mode) and to what extent 7.3 has solved this .. have 2 identical Servers , not sure what the best module ordering may be, so as not to affect / break dependency on a co-located Cilk install (for optimised PHI libs). I suppose a best approach is to install 7.2 / 7.3 on a side by side basis and talk to a third Iperf Server .. OK these are available with ConnectX2 or 3 (i hear X3 degrade IB perf , ( in X2 > X2 // X2 > X3 ) what agnostic OS to mount on the 3rd Server ? and what loading to apply? Certainly like some assist with this.. if u have the time.. even space, should u prefer to have hardware on the bench .. (tho remote install works, why pay 24/7 power bills )
A centos7.3 with the latest Lustre 2.10 is tempting if only to play with the header addressable stack that promises a per-layer vector of function pointers .. finally surfacing after 23 years..
Index of /public/lustre/lustre-2.10.0/el7.3.1611/server/RPMS/x86_64
In this approach, instances of key entity types (files, pages, locks, etc.) are represented as a header, containing attributes shared by all layers. Each header contains a linked list of per-layer `slices',
each of which contains a pointer to a vector of function pointers. Generic operations on layered objects are implemented by going through the list of slices and invoking the corresponding function from the operation vector at every layer.