Since you are trying to upgrade your application server infrastructure, you may want to setup the new environment first using latest version of GlassFish. You could expose the new cluster in the LB tier as a new cluster and then quiesce the old cluster (with a long timeout) to bring it down very gracefully. You may need to manually update the loadbalancer.xml to do this on the LB tier.
re: HADB, GFv2.x supports in-memory. You may try that to see if it satisfies your needs.