Hi, I wrote a simple utility https://github.com/udhos/goben to test TCP throughput between two hosts. At each endpoint, the tool spawns two goroutines: one for reading from net.TCPConn as fast as possible, other for writing into it as fast as possible. However I'm consistently obtaining uneven rates like these:
2018/02/06 23:13:47 report clientReader rate: 20089 Mbps 125557 calls/s 2018/02/06 23:13:47 report clientWriter rate: 417 Mbps 2606 calls/s 2018/02/06 23:13:49 report clientReader rate: 18245 Mbps 114039 calls/s 2018/02/06 23:13:50 report clientWriter rate: 1295 Mbps 8097 calls/s 2018/02/06 23:13:51 report clientReader rate: 19722 Mbps 123263 calls/s 2018/02/06 23:13:52 report clientWriter rate: 2179 Mbps 13620 calls/s 2018/02/06 23:13:53 report clientReader rate: 15851 Mbps 99077 calls/s 2018/02/06 23:13:54 report clientWriter rate: 4274 Mbps 26718 calls/s Notice the input rate (reader) is way higher than output rate (writer). For comparison, this is the result from a single-threaded tool written in C (https://github.com/udhos/nepim): kbps_in kbps_out rcv/s snd/s 3 cur 8 8873837.00 8950252.00 47920.50 34142.50 3 cur 6 8655077.00 8654946.00 47166.50 33016.00 3 cur 4 8773173.00 8772125.00 47800.50 33463.00 3 cur 2 8700428.00 8700690.00 47408.50 33190.50 3 avg 0 8730680.00 8745727.00 47481.90 33362.30 3 min 0 8655077.00 8654946.00 47166.50 33016.00 3 max 0 8873837.00 8950252.00 47920.50 34142.50 I guess I'm hitting some goroutine scheduling issue? How can I improve scheduling fairness (?) in order to get similar loads on both directions? Thanks, Everton -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
