> Doubling a data address as a test address isn't possible in the current
 > Clearview architecture, as you've clearly stated. So I don't see a way
 > out of it.
 > 
 > My objection to using anon groups is that we would have two different
 > schemes in cluster, one for singleton groups, and the other for
 > multi-adapter groups.  It gets messy when the user adds an adapter or
 > removes an adapter and is forced therefore to move between the two
 > schemes.

Good point.

 > Yes. When a group fails, the application gets failed over to a
 > different node.  This usually incur extra outage and users want to to
 > keep the app on the original node as much as possible. So they would
 > typically put two NICs in the group.

I see.

So the core question becomes: is it reasonable to require Sun Cluster
users to configure an additional test address if they want probe-based
failure detection (in addition to the link-based failure detection which
is always available).  If it's not reasonable, then the Clearview team has
some "interesting" engineering challenges ahead of us ;-)

-- 
meem

Reply via email to