Primitive benchmark to measure relative performance with different erlang/rabbit versions and with different options. Only CPU-bound things are measured.
Scenario is very simple: cluster of 3 nodes, a bunch of producers publish to the 1st node, consumers read from the 2nd on. Only normal operation: queues don't grow uncontrollably (so no RAM pressure and no need to do disk swapping), no destructive operations (like killing nodes) occurs during the test.
Despite this simplicity it still could show us what progress was made in rabbitmq/erlang world. And also show the tremendous impact that HA-queues have on performance )
Install apt-cacher-ng on local machine and run:
APT_PROXY_URL=http://10.10.10.1:3142/ ./measure.sh
Without apt-cacher-ng VM provisioning will take substantially longer
and will also create unnecessary network load.
- Different