Thank you for this summary. I was just trying to learn more about it.
A few questions:
1. Do you plan to support external remote dedicated data tier machines such that no state is stored in the app server itself, but consumes data remotely with hotrod or something like this?
2. Would you say that load balancer session affinity is no longer required with DIST? Since hotrod and other protocols already do load balancing for the data tier, would it be possible to remove the HTTP-aware affinity and let the data tier do the intelligent part of the load balancing while the HTTP requests are just sprayed against the available nodes?
3. Do you acquire any cluster locks on the data in the cache during the request processing thread? (like cluster-wide thread synchronization)
4. Do you attempt to collocate the session attribues in the physical cache machine when using ATTRIBUTE-based replication and how? Otherwise isn't there too much effort fetching them separately if needed as bulk? (afaik
https://issues.jboss.org/browse/ISPN-359 is the only plan to do it right now)
5. Is there something new from management perspective? For example would it be possible to read/persistent the configuration though some API or JMX or some other method? May be allow runtime changes without restart to test and adapt different configs depending on traffic and topology changes?