Hello folks,
We have an FS set up where both servers and a portion of clients have
multiple NIDs (one each for IB and eth). I'm having problem where IB clients
will fail over to the eth NID from time to time resulting in a client that has
some OSTs going over tcp0 and some go over o2ib. This is murdering our
performance, as you might imagine. Unfortunately I can't just jettison the
tcp0 NID on the server side as there are legitemate tcp-only clients. Further,
I can't get rid of tcp on the client side because there are other lustre file
systems that mount via tcp... Any ideas how one might control the behavior in
this situation?
----------------
John White
High Performance Computing Services (HPCS)
(510) 486-7307
One Cyclotron Rd, MS: 50B-3209C
Lawrence Berkeley National Lab
Berkeley, CA 94720
_______________________________________________
Lustre-discuss mailing list
[email protected]
http://lists.lustre.org/mailman/listinfo/lustre-discuss
