Quick update: on my home cluster (4 HP 4-core blades, 1GB switch), I
achieved (on a 4 node cluster, 4 senders, udp.xml, everyone sends 5
million 1K messages for a total of 20 million messages) 136'000 1K messages/sec/node !
I'll re-run the same tests as below in the same lab and test
environment a bit later this year and publish some final numbers
here. I assume I'll get even better throughput when I increase
message size from 1K to 2.5 and 5K, as in the tests below !
Okay, done: I re-ran the above tests with 1K, 2.5K and 5K
messages. The throughputs
are 137 MB/sec/node, 150 MB/sec/node
and 153 MB/sec/node ! I assume we get even better perf by
changing to jumbo frames (e.g. MTUs of 8K).