For 1 flow all packets go into one queue and are serviced by a single core, which is limiting the throughput to 1M pps in your case. With multiple flows, the NIC's RSS (receive side scaling) can hash the flows into multiple queues, so the traffic can be processed by multiple cores.
Vasmi ________________________________________ From: discuss <discuss-boun...@openvswitch.org> on behalf of Alexander Shalimov <ashali...@arccn.ru> Sent: Tuesday, September 2, 2014 1:54 PM To: discuss@openvswitch.org Subject: [ovs-discuss] pps throughput depending on the number of flows Hi, We recently finished with throughput measurements of OVS. We saw that phy-to-phy throughput became higher with the increasing number of flows: #1. 1 flow, phy-to-phy, 64 bytes packets => 1Mpps #2. 32 flows (1 src ip, 32 dst ips), phy-to-phy, 64 bytes packets => 9Mpps Could anyone explain why that happens: for 1 flow we have only 1Mpps, but for 32flows we have 9Mpps? It seems to be related to multicore and concurrency. But where is this code located (flow scheduling and etc)? Thanks and Regards! -- С уважением, | Sincerely, Александр. | Alexander. _______________________________________________ discuss mailing list discuss@openvswitch.org https://urldefense.proofpoint.com/v1/url?u=http://openvswitch.org/mailman/listinfo/discuss&k=oIvRg1%2BdGAgOoM1BIlLLqw%3D%3D%0A&r=DZ2AGN9X6konlUBIZhMbNm3P4FshcGkQMzt4v3zsGQk%3D%0A&m=%2B7Iui67wuFdmgL43hB71qChkMhfwqV86AEf31DEOP00%3D%0A&s=54295a67ba1ee01e9e1c533d8836ec5d732f9bf4660b906dddde655281914180 _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
