In a cluster of two brokers, running on two servers, should I also run the same ESB container twice, on each server? I guess this is obvious: If one server goes down, the other instance of the ESB container should be available to handle the processing.
Indeed you are correct. You may also be able to achieve better through put as the load of the ESB processes can be shared across the two servers.
We've taken a different approach. We setup two different MF containers and deploy them on different servers. ctSvc1-Prod and ctSvc2-Prod
We then deploy the same ESB container instance in each MF Container. We use Topics so we just set the "Shared Subscriptions" property to true on the ESB endpoint configuration.
This allows the two ESB containers running in two different MF Containers to process messages through a round-robin fashion without processing the same message twice. This way we can scale as many times as required by adding additional MF containers and deploy the same ESB container in each of them respectively...
We get High availability, but still have to take worst case into account. When we figure one server cannot handle worst case load we will deploy another MF container and have three instances of our ESB processes running. In best case we have three servers processing messages and hopefully worst case we will have two. Just depends on how high - "High availability" you want.