I'm capping out at around 140,000 msg/s that's with my little javascript wrapper, https://github.com/reqshark/nanomsg.iojs on my laptop, msg latency below measured in JavaScript Date.now() milliseconds ✘-130 *bent**@**quad* */Users/bent/nmsg/nanomsg.iojs * [*master*|● 1✚ 4] 02:23 $ node v8 msg count: 10000, msg latency: 108 msg count: 20000, msg latency: 192 msg count: 30000, msg latency: 288 msg count: 40000, msg latency: 349 msg count: 50000, msg latency: 413 msg count: 60000, msg latency: 496 msg count: 70000, msg latency: 549 msg count: 80000, msg latency: 606 msg count: 90000, msg latency: 701 msg count: 100000, msg latency: 752 msg count: 110000, msg latency: 848 msg count: 120000, msg latency: 904 msg count: 130000, msg latency: 956 msg count: 140000, msg latency: 1056 msg count: 150000, msg latency: 1117 msg count: 160000, msg latency: 1173 msg count: 170000, msg latency: 1254 msg count: 180000, msg latency: 1322 msg count: 190000, msg latency: 1399 msg count: 200000, msg latency: 1451 msg count: 210000, msg latency: 1506 msg count: 220000, msg latency: 1582 msg count: 230000, msg latency: 1661 msg count: 240000, msg latency: 1709 msg count: 250000, msg latency: 1786 msg count: 260000, msg latency: 1834 msg count: 270000, msg latency: 1947 msg count: 280000, msg latency: 1994 msg count: 290000, msg latency: 2043 msg count: 300000, msg latency: 2132 msg count: 310000, msg latency: 2183 msg count: 320000, msg latency: 2232 msg count: 330000, msg latency: 2316 msg count: 340000, msg latency: 2370 msg count: 350000, msg latency: 2449 ^C ✘-130 *bent**@**quad* */Users/bent/nmsg/nanomsg.iojs * [*master*|● 1✚ 4] On Wed, Jan 21, 2015 at 1:45 AM, junyi sun <ccnusjy@xxxxxxxxx> wrote: > I think 50000 msg/s is good enough. I used to make performance test on > Redis and memcached. Redis can reach 72000 msg/s, memcached can reach 25000 > msg/s. > > The speed of request/reply pattern is limited by the round trip cost of > TCP. If we want much higher qps, I think we should use asynchronous > pattern, in which the users can register a callback function for request > and pick the corresponding response when it arrived. > > > > On Tue, Jan 20, 2015 at 4:03 PM, Pierre Salmon < > pierre.salmon@xxxxxxxxxxxxx> wrote: > >> For information, I already implemented this example and i obtained only >> 50000 msg/s. >> >> Pierre >> >> >> On 01/20/2015 03:37 AM, Garrett D'Amore wrote: >> >>> socket used by the worker. That means you have to save the header and >>> restore it — the device() routine has this logic, but you need to copy that >>> logic as appropriate, rat >>> >> >> >> > -- Bent Cardan nothingsatisfies.com | bent@xxxxxxxxxxxxxxxxxxxx