=============================== #opnfv-vswitchperf: vswitchperf =============================== Meeting started by SridharRao at 15:02:00 UTC. The full logs are available at http://ircbot.wl.linuxfoundation.org/meetings/opnfv-vswitchperf/2018/opnfv-vswitchperf.2018-06-27-15.02.log.html . Meeting summary --------------- * Development Update (SridharRao, 15:02:14) * Xena-Patch from Christian/Bob multi-flows is merged. (SridharRao, 15:03:10) * Logging Support VSPERF - Should Timestamps be added? (SridharRao, 15:04:01) * Timestamps should be added. (SridharRao, 15:06:38) * Pod-12 Status (SridharRao, 15:07:16) * Al Morton (acm, 15:07:20) * Intel Pods IP (new) We are back online at 192.55.66.214 with the exception of traffic generator. (acm, 15:08:43) * Rerun Tests with Timestamps enabled in logs. (SridharRao, 15:11:42) * VSPERF-Intel Demo IEEE SDN-NFV (SridharRao, 15:13:23) * Create Cache Profiles - L3 Cache. (SridharRao, 15:14:02) * Find a test VNF that is cache-hungry [ Data-plane VNFs - AVB, Video-Servers, TGen-10G-64B] (SridharRao, 15:14:54) * 35-55MB of L3-Cache per socket (SridharRao, 15:17:18) * Experiment Test-VNF and Noisy-Neighbor - without dedicated cacheways (SridharRao, 15:18:51) * Repeat experiment with dedicated cacheways (SridharRao, 15:19:04) * OS Nova cannot balance load across NUMA Nodes (Alec) (acm, 15:23:34) * No OS on Pod 12 (acm, 15:23:56) * Need Pod 19, where there will be a full OS and OPNFV Deployment, then NFV Bench (acm, 15:24:44) * Alec - multiple instances of Testpmd may be sufficient to consume cache (acm, 15:26:00) * Container-Networking - Vswitch-Based Only ? (SridharRao, 15:26:32) * Many options, and K8s adds even more networking options! (acm, 15:27:35) * Sridhar - Should we focus on vSwitch scenarios? (acm, 15:28:32) * Alec - first decision is to use K8s or not (acm, 15:28:51) * SR - Assume no K8s, just Docker Containers, is this worthwhile place to start? (acm, 15:29:45) * vSwitch would be running on baremetal (acm, 15:30:19) * Christian - not plugged-in yet, need to ask RH perf team, get some input (acm, 15:31:05) * (1) Presentation on VSwitch-Based Container Networking. (SridharRao, 15:31:15) * Need info on common use cases today (acm, 15:32:18) * Kubernetes + vSwitches: Solutions/Studies existing (SridharRao, 15:32:35) * Alec - K8s only understands L3, is IP based, no L2 with vSwitch, SR/IOV is working for containers, too. (acm, 15:34:08) * Alec - Intel and VPP/FD.IO have studies on Kubernetes+vSwitch (SridharRao, 15:34:51) * Alec - K8s uses TCP/IP stack in host, and there are some scaling issues when using many containers (acm, 15:35:34) * #info Alec - experiments running TCP/IP stack in user space, with VPP?, with container connectivity at L3 (acm, 15:37:09) * ACTION: Alec627 to send FD.io doc on container networking (acm, 15:40:31) * acm agree with this path, also learn from what others have done when measuring in this space. (acm, 15:44:25) * Alec - L3 today in K8s, but this is not efficient when connecting multiple containers. (acm, 15:51:49) Meeting ended at 15:58:08 UTC. People present (lines said) --------------------------- * acm (19) * SridharRao (18) * collabot_ (3) * marst (1) Generated by `MeetBot`_ 0.1.4