Rumor has it that on Wed, Jan 28, 2004 at 05:16:51PM -0500 James Bottomley said: > On Wed, 2004-01-28 at 16:02, Smart, James wrote: > > I'd add one additional issue : > > > > 3) SCSI Reservations > > > > I guess this could be considered a part of (2), but given that reservations > > bring in a host of additional problems, and potentially brings up multi-host > > configs/clusters as well... > > The current crop of cluster implementations that use reservations manage > quite well in user space alone. Certainly, though, the presence of > reservations can interfere with failover code and may be considered a > problem that needs to be solved in multi-path...it all depends on how > the multi-path devices react to path switching in the face of > reservations. > It needs to be known to the pathing layer if you've got load balancing. It has to know which path has the reservation and only use that one. > I'm open to suggestions on this, but I think it should be done > separately as part of a device ownership API. > > James > -- Philip R. Auld, Ph.D. Egenera, Inc. Principal Software Engineer 165 Forest St. (508) 858-2628 Marlboro, MA 01752