I have a non-hypothetical concern re: FD usage with ZMQ. Admittedly, I am a typical programmer having implemented most inter-proc comm via bare sockets.
As I understand the general use-model, a program would would use multiple TCP ports for the various message-types to be sent between servers and clients, PUB/SUB, REQ/REP, PUSH/PULL. That being the case, comm between two procs, a server could easily use 2 or more sockets for each client. If a proc is limited to 5k FDs then the number of clients simultaneously connected is limited to less than 2.5k before it has to start managing those FDs with some kind of LRU algorith with reconnects as necessary. That seems to be the inevitable outcome of not being able to map an endpoint to multiple sockets. Is that correct? What's the solution? Is that actually acceptable? Thank you. FWIW, ZMQ seems very promising to use in my current project. I have two key concerns re: it; this is one of them. CZC _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
