Brian, > So there may be a place for a pure-ruby implementation of zmq after all.
I would definitely try not to go that way. While the wire protocol is extremely simple, actual implementation of a messaging endpoint is pretty complex (managing many connections, queueing, fair queueing and load balancing etc.) Moreover, pure Ruby implementation is bound to be slow. The main problem here (and that's a reoccurring issue) is the 0MQ sockets cannot be polled on because they aren't real file descriptors. I can imagine a small wrapper library on top of 0MQ that would intercept the system BSD socket API and forward the calls either to OS (in case of OS file descriptors) or to 0MQ (in case of 0MQ sockets). OFED's implementation of SDP does exactly the same thing, so it should be feasible. The question is how would that work on Windows... In any case tcmalloc seems to be able to intercept malloc even on Windows so there should be some way. Martin _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
