Switch testing | ||||||
| posted by Brian Krusic on March 25, 2015, 1:55 p.m. | ||||||
| ||||||
| Thread Tags: discuss-at-studiosysadmins | ||||||
| ||||||
| Hi, I suggest doing netperf as well. I recently used these to debug my 40Gb network, special thanks to Richard Hastie@Mellanox BTW, Iperf for max BW i typically use : Server: iperf -s -l64k Client iperf - c<ip of server> -l64k -P2 Netperf Server: netserver Client: netperf H <server IP addr> t TCP_RR The result is transactions per second (TPR). Latency is calculated as: Latency=(1/TPR)/2 netperf -H 11.10.10.12 -t TCP_STREAM -I 99,5 -T 2,2 Another few examples netperf H <server-ip> -t TCP_STREAM I 99,5 T 2,2 netperf H <server-ip> -t TCP_STREAM I 99,5 T 2,2 -- -m 2048 netperf H <server-ip> -t TCP_STREAM I 99,5 T 2,2 -- -m 1024 netperf H <server-ip> -t TCP_STREAM I 99,5 T 2,2 -- -m 500 netperf H <server-ip> -t TCP_RR I 99,5 T 2,2 -- -m 1024 netperf H <server-ip> -t UDP_STREAM I 99,5 T 2,2 netperf H <server-ip> -t UDP_STREAM I 99,5 T 2,2 -- -m 1024 netperf H <server-ip> -t UDP_RR I 99,5 T 2,2 -- -m 150 Of course tweak your server/client cards as well and not just TCP fromage but IRQ affinity etc - Brian Technologist... i.e., Purveyor of the ironic" On Mar 25, 2015, at 10:31 AM, Jathavan Sriram <sriram@harvest-postproduction.com> wrote:
|