I''m hoping somebody may have some experience in this arena. We''re evaluating several vendors of storage-array products. Many of them run FC as the host-side connection, using SCSI protocol over that to control the device. One issue that''s been popping up in our evals is that some of the devices have characteristics which seem like they''d cause problems for lustre failover. For instance, I did some testing yesterday in which it developed that I can configure up a lun and make it available over N of my host ports, and everything''s happy. But when I scan luns from the hosts, what happens is that whenever a host probes at a lun, the controller essentially thinks a failover is happening, and swaps the lun over to the port the probe came in on. None of this would be an issue if I weren''t planning on using lustre''s failover capability, because it would be one-to-one, but if I propose to use lustre failover, it seems like I need to either find ways of reconfiguring the array simultaneously with the failover, or find a way to make the host not do its scan until the failover happens. Both of which seem dodgey. Comments from the vendors, and other things I''ve seen in discussions about FC/SCSI, have led me to believe that this whole area is not completely explored territory, but I''ve been out of the SCSI game for years and haven''t kept track of what''s the current state of the art. If anybody has any insight, or can supply a description of real-world configurations which solve this problem, I''d appreciate hearing it. TIA... John Dunning http://www.sicortex.com Dense Cluster Computing