From: Dalesio, L. <da...@bn...> - 2013-04-10 11:28:56
|
I have to admit that performance measurements are my favorite things. Will you be presenting these numbers at the EPICS meeting? Bob ________________________________ From: Matej Sekoranja [mat...@co...] Sent: Wednesday, April 10, 2013 7:08 AM To: White, Greg Cc: epi...@li... Subject: Re: First successful transport of Area Detector image to client running plugins I've updated testGetPerformance utility to output network throughtput. My setup is: - testServer running on MacBook Pro 2.66GHz Inter Core Duo, 8GB RAM, 1Gbit network card - 1Gbit HP switch - testGetPeformance on Intel Core i7-3820 CPU @ 3.60GHz (4C/8T) So, we are limited by 1Gbit network. > testGetPerformance -s 10000000 -i 10 -l 0 1 channel(s) of double array size of 10000000 element(s) (0==scalar), 10 iteration(s) per run, 0 run(s) (0==forever) all connected 6.888271 seconds, 1.452 (x 1 = 1.452) gets/s, data throughput 0.865 Gbits/s 6.851985 seconds, 1.459 (x 1 = 1.459) gets/s, data throughput 0.870 Gbits/s 6.846615 seconds, 1.461 (x 1 = 1.461) gets/s, data throughput 0.871 Gbits/s 6.847903 seconds, 1.460 (x 1 = 1.460) gets/s, data throughput 0.870 Gbits/s 6.853981 seconds, 1.459 (x 1 = 1.459) gets/s, data throughput 0.870 Gbits/s 6.855381 seconds, 1.459 (x 1 = 1.459) gets/s, data throughput 0.869 Gbits/s So, it's 0.87Gbit/s compared to the iperf TCP benchmark: > iperf -c 192.168.1.112 ------------------------------------------------------------ Client connecting to 192.168.1.112, TCP port 5001 TCP window size: 23.5 KByte (default) ------------------------------------------------------------ [ 3] local 192.168.1.110 port 40558 connected with 192.168.1.112 port 5001 [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec 1.09 GBytes 937 Mbits/sec So, it's 0.937Gbit/s vs. 0.870Gbit/s. Not bad, if you think C++ I haven't spent time on optimized code. If I do all local test on i7 machine I get 14Gbit/s vs 40Gbit/s. That's something I'll am eager to improve. Matej |