Hi Luca, thanks for your reply.
2017-08-10 21:28 GMT+02:00 Luca Boccassi <luca.bocca...@gmail.com>: > On Thu, 2017-08-10 at 19:44 +0200, Francesco wrote: >> thanks for details. I found out a problem that was blocking my >> testing: I have used a gcc 5.3 on some servers to build Zyre and >> apparently some build option I have there is causing my problem >> (nodes >> stuck after "Took 3 ms to coordinate with all remote")... in another >> couple of servers I used the Centos7 native compiler (gcc4.7 IIRC) >> and >> it worked. I will dig into this issue tomorrow. > > That's a bit suspicious :-) > Actually I found that the utility "zpinger" works (the 2 nodes "ping" each other)... I suspect it's due to the fact that when I used gcc5.3 I used a version of libzmq built without libsodium... >> When I got it running however I got somewhat low numbers, around 60k >> msg/sec (2 nodes). On the same machines I ran my own performance >> testing program with ZeroMQ and for 64B packets I reached around 1.5 >> Mpps. >> >> I didn't check the source code of the perf_loca/remote program but am >> I missing something? > > I get the same numbers more or less. Note that zyre perf code is less > optimised for throughput - it uses high level APIs that do a loc of > mallocs and copies, etc. > libzmq perf code uses more high performance zero-copy APIs, and that > makes a lot of difference. So do you think that when I have 2 Zyre nodes talking to each other (actually unidirectional communication is a more fair comparison to my ZMQ tests) I can reach similar performance to plain ZMQ sockets if I spend some time optimizing the code? That's an important point for me to decide whether using vanilla ZMQ or Zyre instead... Thanks, Francesco _______________________________________________ zeromq-dev mailing list zeromq-dev@lists.zeromq.org https://lists.zeromq.org/mailman/listinfo/zeromq-dev
