[rabbitmq-discuss] [Platform Team] Preview Release of Haigha

Mon May 16 13:34:06 BST 2011

Gavin,

Thank you, I knew there was a good reason for kicking this out the
door despite some missing tests, documentation and features.

These numbers are very interesting, I'm most surprised by the single
channel one. I'll take a look at that. The test script supports
running a profiler, which I've used to fix a lot of bottlenecks.
That's why I picked the 500 number, but also just because I like to
push network protocols as hard as I can.

I wouldn't be surprised if the depth of the stack or the time-slicing
in the underlying socket abstraction incurs an overhead at one channel
that goes away once more data is flowing. We could also test with a
single channel but more normal conditions, where that channel is
trying to consume from a queue under heavy load, or push large volumes
of data.