14:59:04 #startmeeting VSPERF weekly call 14:59:04 Meeting started Wed Sep 28 14:59:04 2016 UTC. The chair is mtahhan__. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:59:04 Useful Commands: #action #agreed #help #info #idea #link #topic. 14:59:04 The meeting name has been set to 'vsperf_weekly_call' 14:59:44 #info Roll Call 14:59:50 #info Maryam Tahhan 15:01:28 #topic Agenda 15:01:38 #info dev update 15:01:46 #info QoS Readout 15:01:57 #info Release D 15:02:05 #info AOB 15:02:28 #topic dev-update 15:05:32 #info bug fixes worked on as well as cli parameter updates 15:05:39 #link https://gerrit.opnfv.org/gerrit/#/dashboard/self 15:05:52 #info don't be afraid to merge +1'd code 15:06:06 #info Moongen - latency testing is being worked in 15:06:20 #info Moongen - running in a VM - work ongoing there 15:07:17 #info as part of this effort - virtio drivers from master moongen branch have been merged with Bill's tree 15:07:31 #info not published yet 15:08:33 #info Bill is seeing packets getting dropped on the Receive side of the virtio ports into the VM on both his branch and the master moongen branch 15:10:28 #info don't know if the moongen maintainers are aware of the latency bug that exists 15:10:41 #info github issue will be opened against the repo 15:12:28 #info 2889 address caching and learning tests still under testing 15:12:48 #info Sridhar looking at Yardstick support from the yardstick. 15:13:40 #topic QoS Testing Readout - Sridhar 15:15:02 #action - Sridhar to send link to "An internet-wide analysis of Traffic policing" paper to the mailing list 15:15:32 #action Sridhar to upload readout to the readouts wiki 15:15:51 #info Sridhar walking though how to setup QoS for OVS 15:16:27 #info OVS doesn't implement QoS itself 15:16:48 #info it uses a subset of Linux Kernel QoS features 15:17:32 #info Policing works as expected 15:17:56 #info traffic shaping is a possible test case that we could look at benchmarking 15:23:16 #info Al points out that if OVS vsctl is using tc to configure the queuing - then the queuing is happening on the Linux Plane not the vswitch plane 15:24:14 #info then the testing would really be of the Linux Kernel Queue 15:24:55 #info Al points out that there is a traffic management benchmarking rfc - 7640 that we might be able to leverage 15:27:16 #link https://tools.ietf.org/html/rfc7640 15:27:36 #topic Release D planning 15:34:38 #info goals for moongen - latency testing, CI integration 15:35:04 #info Al suggests that maybe RFC5481 can be implemented with moongen 15:35:23 #info Q. if there's a limit on samples, how many samples can you get? 15:35:35 #info 800/900 per second 15:35:47 #info it may not be enough 15:36:12 #info adding more options for tuning to vsperf 15:36:57 #info binding options, profiles, moving to libvirt instead of Qemu, adding testpmd 16-07 as a switch 15:37:09 #info are some options 15:37:38 #info vsperf integrated in a more openstack env 15:39:15 #info Yardstick integration. 15:41:50 #info IMIX testing is also something that could be implemented for D 15:44:34 #info 40Gb integration is also an option for D 15:47:49 #info Netronome would like to contribute back to vsperf the following features: NIC integration for 40Gb and 25 Gb. Moongen support for Netronome NICs, Drop rate tolerance and traffic patterns 15:48:27 #info Moongen changes are in the master moongen branch 15:48:56 #info compilation changes to DPDK for the NICs aren't upstreamed to Moongen yet 15:53:35 #action Bill to look Moongen feature planning for D next week 15:55:08 #action Maryam to to the same for yardstick 15:55:32 #info Christian - vsperf tuning and options 15:56:05 #info Prevaze - Netronome features 15:56:42 #info review IMIX stories already in JIRA 15:57:02 ixia chassis in intel-pod3 should be working again 15:57:36 #endmeeting