Hi all, I have committed NetBSD support to git. This includes support for atomic_ops(3) if you are using NetBSD 5.0 or newer and have fixed PR 42842 which I have reported upstream. This should be fixed by default in the next NetBSD release (5.0.3).
Also note that due to NetBSD not supporting AI_ADDRCONFIG if you have IPv6 and IPv4 addresses configured on an interface bind to IPv4 you will have to specify the numeric IP address in calls to zmq_bind(). Some performance figures to give an idea of what can be expected on vintage hardware: These tests were run on a HP Jornada 728 with a 206Mhz SA-1100 processor, with local_thr running on the Jornada and remote_lat on another (faster) machine on a 100mbps LAN. The Jornada is connected to the LAN using a 16bit 3COM PCMCIA NIC using the ep driver. The test was done using TCP. I do not include tests for local loopback as running both ends on the Jornada just results in the entire test being CPU-bound so the results are not useful. with atomic_ops(3) msg size msg count msg/s Mb/s 1 500000 104442 0.836 4 500000 81088 2.595 8 500000 64750 4.144 16 500000 45402 5.809 32 500000 21959 5.622 48 500000 17017 6.535 64 500000 13864 7.098 without atomic_ops(3) msg size msg count msg/s Mb/s 1 500000 86245 0.690 4 500000 70849 2.267 8 500000 57219 3.662 16 500000 41445 5.305 32 500000 20240 5.181 48 500000 15993 6.141 64 500000 13154 6.735 The drop at size=32 is due to ZMQ_MAX_VSM_SIZE == 30. Not half bad for hardware of this class. I heard somewhere that you can run an entire stock exchange on 20k messages/s :-) Cheers, -mato _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
