#opnfv-cperf: cperf

Meeting summary

1. 1. dfarrell07 is at OPNFV plugfest with CPerf hat on (among other hats) (dfarrell07, 18:14:16)
   2. dfarrell07 is Working with Pan of SDNCTC on collab. Trying to get access to the perf testing tool source they use in their reports. (dfarrell07, 18:14:51)
   3. ODL perf whitepaper 1.1 is almost done, about to be sent to ONOS for review and then publication (dfarrell07, 18:15:23)
   4. Basically they released 1.5.1 with NB REST batching and stats-colleciton related perf fixes, want us to retest (dfarrell07, 18:17:28)
   5. Disagreement about testing methods, they use interal Java APIs and we think that's not useful for users, just devs (dfarrell07, 18:17:38)
   6. dfarrell07 Talked to jack about OPNFV POD 2, should be ready for our use (dfarrell07, 18:17:54)
   7. dfarrell07 continued to play with NSTAT a bit, not as much as I hoped (dfarrell07, 18:18:08)
   8. jamoluhrsen has continued with our hello world job, patch merged (dfarrell07, 18:18:56)
   9. our docker image isn't on DockerHub yet, so that'll fail until fixed (dfarrell07, 18:19:08)
   10. https://build.opnfv.org/ci/computer/intel-pod2/ (jamoluhrsen, 18:19:44)
   11. POD 2 is broken, reports jamoluhrsen (dfarrell07, 18:20:27)
   12. ACTION: dfarrell07 will follow up with Jack (dfarrell07, 18:20:34)
   13. https://build.opnfv.org/ci/computer/intel-pod2/ broken pod info (dfarrell07, 18:20:50)
   14. Nikos has continued with NSTAT perf testing, both for Li and Be (dfarrell07, 18:21:26)
   15. Be is better than Li in end-to-end flow perf for all topos and sizes, in extreme cases where Li fails Be wins. 1m flows and 5k switch topos for example. Be is making progress! (dfarrell07, 18:22:07)
   16. Nikos will likely publish updated PDF with these results early next week (dfarrell07, 18:22:23)
   17. Nikos is working on tool improvements, things releated to var size pkt ins (dfarrell07, 18:22:39)
   18. Testing above has to do with He OFP, not newer Li one (dfarrell07, 18:23:12)
   19. Previous intercom telco report has Li OFP design results for past tests, but went with "default" this time (dfarrell07, 18:23:55)
   20. Li doing better than Be on some NB REST tests, ignoring SB parts, just the RESTCONF perf (dfarrell07, 18:24:59)
   21. https://drive.google.com/file/d/0B_7IPKiQjuIZTkh5SVBkaWJvX3c SDNCTC perf report (dfarrell07, 18:36:15)
   22. Jamo points out that it's possible to write high-perf apps that exist inside the controller and would use the Java APIs, so we should be careful about saying that apps would only use NB REST APIs (dfarrell07, 18:37:51)
   23. abhijitkumbhare joined today, mostly to watch, welcome! (dfarrell07, 18:39:08)

Action items

1. dfarrell07 will follow up with Jack

Action items, by person

1. dfarrell07
   1. dfarrell07 will follow up with Jack

People present (lines said)

1. dfarrell07 (26)
2. collabot` (4)
3. jamoluhrsen (1)