The clustering docs don't go into too much detail about the load balancing algorithms..
I have the following setup: A client program with 6 threads running sending stateless session bean calls to a cluster of 4 nodes. Each call takes a node about 1 second to complete.
Now the problem with roundrobin is that the slowest node in the roundrobin makes everybody else wait. When I switch to FirstAvailable, then either:
a) one node grabs all the calls, which is useless because it doesnt add parallelism b) the nodes seem to do roughly the same amount of work, but at the end of say 1000 calls, the slowest node seems to process about 50 by itself, i.e. it has somehow "collected" a large set of calls..
Has anybody managed a proper setup with firstavailable that can handle a scenario of different spec machines?