Hi Paul, for PUB->SUB and PUSH-->PULL I will use inproc or ipc and only req->reply is used via tcp. I am playing around currently to get some feeling for nanomsg. Stacking of the patterns is something which is suggested by zmq as far as I understood. This might be the reason why I am sticking to it. I am trying to leverage from all patterns somehow the best: from PUB/SUB the subscriptions, req->reply for ackn and push-pull for load-distribution to worker-nodes... Best, Joerg 2013/11/22 Paul Colomiets <paul@xxxxxxxxxxxxxx> > Hi Jörg, > > > On Fri, Nov 22, 2013 at 6:33 PM, Jörg Singler <joerg@xxxxxxxxxxxx> wrote: > > I used the following stacking > > of patterns: Pub -> SUB (filtering on topics) -> Files > > I'm not sure about your use case, but pub-sub chain can loose > messages when pushback is applied (when subscriber is not fast > enough). And I'd say this defeats the whole purpose of the > persistence. > > > -> req -> reply > > (sending aknowledge only) -> PUSH --> PULL (distributing work to multiple > > workers). > > I don't think you need aknowldedge here, because you can loose a > message in push-pull anyway. Unless you do push/pull via inproc/ipc. > > > Do you think the stacking should/can be optimized? I will keep you > > updated... > > > > I think you are thinking in terms of "lets make rabbitmq over nanomsg" > rather than "lets solve the task in neat way". I.e. you use 3 message > patterns for doing a single task. That's not how nanomsg is supposed > to work IMO. > > Do you have higher level task in mind, that you can describe? Or do > you just playing with things as time permits? > > -- > Paul > >