On Feb 9, 2012, at 7:42 AM, Calvin de Vries wrote: > I'm not sure I agree that ZMQ_AFFINITY should be modified at all. I think > that it would be most beneficial if you could still use ZMQ_AFFINITY so that > a socket is on a particular I/O thread that has already been pinned to a > specific CPU.
He wasn't proposing how ZMQ_AFFINITY works. Any change would be to accommodate the arbitrary-length bitmask used for assigning I/O threads to CPUs. There are already machines available that have *more than* 64 CPUs or cores, so the uint64 used by ZMQ_AFFINITY could only address up to 64 I/O threads. I can't imagine a case right now where it would be necessary to do so, but then again 640k of RAM is enough for everybody, right? :) cr _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
