[zeromq-dev] taskvent/taskwork2/tasksink2 performance degredation

[Pieter Hintjens]

On Wed, May 18, 2011 at 8:43 AM, John D. Mitchell <jdmitchell at gmail.com> wrote:


>>> Ian wrote:
>> Is it possible there are old sockets still handing around from previous runs when you do the later ones?
>
> I've run the tests quite a few times now with different numbers of workers, etc. and with variety of pauses between tests. I haven't noticed any oddly lingering sockets. All of the workers and the ventilator and sink all exit correctly -- just the performance oddity.

My first guess would be the same as Ian's, namely that there are
processes remaining alive after each round.

The number of messages is extremely low, and all the processes do is
sleep a while before pushing the message out. There is a 0%
performance aspect this example, it's purely a demonstration of
pipeline synchronization.

John, I'd suggest (seriously) that this is an excellent place to learn
how to debug 0MQ flows. You don't mention what language you're working
in, but add print statements to each node, when it sends and receives
a message, and use these to understand precisely what's going on. You
will sooner or later see the problem, and then tell me what the
examples do wrong. :-)

-Pieter