> > There would be two ways I think:
> > * dedicate each node to a particular entity -> no need
> > to update
> > * Use a CMP implementation that does synched cluster
> > caching. Use, for example, timestamps for optimistic
> > concurrency control, or global locks for pessimistic CC
>
> > The latter /w OCC would be the more likely case I guess.
>
> The former precludes object failover, which is the primary reason for
> clustering in the first place.
No, it would not. If the node running the entity failed the cluster
could fail over to use another cluster node.
> The latter requires so much overhead (and serialization) that any benefit
> derived from Option A (entity caching in the container) is lost.
It is very possible to implement a synched cache with minimal overhead
and *no* serialization (you are making way too many assumptions about
the implementation here), hence making Option A caches very nice for
performance, especially in read-mostly scenarios (no db access at all
for extended periods of time).
I will demonstrate this in source in jBoss later on.
/Rickard
===========================================================================
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff EJB-INTEREST". For general help, send email to
[EMAIL PROTECTED] and include in the body of the message "help".
