We could dynamically inject a custom instance of a DISCOVERY protocol in the JGroups stack which doesn't require additional configuration from the end user to get both Infinispan and JGroups stacks working out of the box in any environment (including EC2 or other cloud systems lacking multicast support).

Hibernate Search users are supposed to configure a shared database anyway so that would be a reliable shared storage for network information, we just need to define (and use) and additional table in the DDL generation.