> Marcin Gozdalik wrote: > > > Avg. us/call Elapsed Name > > 0.064833 0.648334 clock_gettime(2/CLOCK_REALTIME) > > 0.009635 0.096351 clock_gettime(2/CLOCK_REALTIME_COARSE) > > That's a pretty good performance! I wouldn't hesitate to use it in > zmq_poll. The impact on non-Linux systems is still unknown though.
Still, the solution is not obvious. 9 or 64 nanoseconds to measure time as shown above would be more or less OK. It would decrease throughput in tight zmq_poll-based loops by some 5-20 per cent. However, if some OSes report time measurement taking 10 ms it would decrease throughput 5000 times or so -- which is not acceptable. Martin _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
