On 01/27/2010 10:12 PM, Lars Marowsky-Bree wrote: > On 2010-01-27T17:51:15, Florian Haas <[email protected]> wrote: > >> And if you want OCFS over 3 nodes _and_ DRBD redundancy, create a highly >> available iSCSI target (using regular Single-Primary DRBD), then mount >> that iSCSI target from 3 nodes. > > Could you run drbd in dual-primary mode, export via iSCSI from both > primaries via iSCSI, and use those iSCSI devices on the "clients" to > mount OCFS2 on?
Let's stick to SCSI parlance and refer to the "clients" as initiators. As they may well be (and usually are) servers themselves. :) Theoretically you could do that. In such a setup, you could conceivably create individual per-node targets, listening on per-node IPs, and then use a multipath I/O setup on the initiators. But whether you achieve any actual benefit from that is quite doubtful. I currently don't know of any iSCSI implementation that is capable of creating or enforcing cluster-wide Persistent Reservations, for example (not that that's a concern for the OCFS2 use case, but you get the point). I also wouldn't want to use round-robin load balancing across these multipath links. Bottom line: while this may be possible, the production user base for this setup, to the best of our knowledge, is currently zero. Feel free to try it out and report back your results. :) Cheers, Florian -- "In theory, that should be impossible" - Lars Marowsky-Brée, #linux-cluster, January 2010
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
