>
> However, by making sure the two processes doesn't change CPU during
> runtime, sockets are always the fastest, and I get consistent results:

That is useful. How consistent are the results people are getting though? When
I run with 1M messages the results are generally very flakey, 2, 4, 6 seconds
between different runs. If I use 10M then due to the far longer run time then
the differences are will 2 to 6 seconds but the run time is a whole order of
magnitude different give a different picture.

Will try and find an hour to run the tests with this CPU selection code.

Regards, nick.