On 06/01/2011 10:40 PM, Pieter Hintjens wrote: > Right. The problem in our use case is that the server will send > messages much faster than the network can handle them, and certainly > faster than most clients, so full queues will be a normal situation.
I am not sure what the actual use case is, but what I infer is that the clients are not able to receive replies as fast as they produce requests (for example because the replies are huge while the requests are small). Consequently, replies are being dropped not to run out of memory which lowers the reliability of the system as a whole. If that's the case, we could try to introduce "retroflex pushback" ie. we can stop reading requests from a connection that has the reply queue full. That way the client will be forced to read the replies before sending more requests. Martin _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
