#opendaylight-sfc: weekly ODL SFC meeting

Meeting summary

1. 1. Brady Johnson (ebrjohn, 16:59:39)
   2. https://lists.opendaylight.org/pipermail/sfc-dev/2016-November/003697.html agenda (ebrjohn, 17:00:14)
   3. Diego Granados (dgranados, 17:00:22)
   4. Sam Hague (shague, 17:04:57)
   5. Brady talking about the context of this meeting. The implications of implementing SFC not on NSH, but on VXLAN (edavsua, 17:06:38)
   6. Ericsson has plans to use plain VXLAN, but not abandon NSH (edavsua, 17:08:15)
   7. we need to re-classify, since we are not going to have the NSH metadata at hand (edavsua, 17:08:59)
   8. VXLAN has some scalability limitations, but it is widespread (edavsua, 17:09:30)
   9. We can use a flag on the SF model (edavsua, 17:10:00)
   10. Ericsson will continue supporting NSH, this is just a workaround until we have "official" support in OVS (edavsua, 17:10:37)
   11. M2 milestone is comming in a couple of weeks. So we need to have a plan in place for that date (edavsua, 17:11:24)
   12. Jaime, Diego and the rest of the Ericsson's team are working on the integration of SFC with Genius (edavsua, 17:12:28)
   13. The integration with Genius is done in the context of the Logical SFF (edavsua, 17:13:13)
   14. Jaime tells we are just finishing the feature, even the classifier integrated with Genius (edavsua, 17:13:44)
   15. We still need to migrate the sfc-genius component to blueprint (edavsua, 17:14:15)
   16. we haven't done any major changes in the current model of the classifier (edavsua, 17:14:52)
   17. the classifier finds out the neutron port for the data plane node where to send the traffic (edavsua, 17:16:04)
   18. the model remains the same (edavsua, 17:16:17)
   19. Sam, how are we going to reclassify using the VXLAN? (edavsua, 17:17:28)
   20. https://lists.opendaylight.org/pipermail/sfc-dev/2016-November/003625.html Daya's response where she says SFC can remove the NSH header (ebrjohn, 17:19:23)
   21. Jaime: the classifier is a higher priority service that comes after SFC (edavsua, 17:20:42)
   22. Sam: Genius will handle the metadata doing the same type of work we are doing noe (edavsua, 17:21:17)
   23. Jaime: can the classifier handle the use case in which the VM migrates (edavsua, 17:21:59)
   24. Jaime: we are working on that now in the context of the Logical SFF (edavsua, 17:22:34)
   25. Sam: right now we are adding all the rules needed on all the nodes where the client might migrate to (edavsua, 17:24:45)
   26. Jaime: does this mean that the Classifier is located always in the first SF? (edavsua, 17:26:40)
   27. We have a VXLAN-GPE on every node right now (edavsua, 17:27:27)
   28. the client should be outside the SFC domain (edavsua, 17:27:48)
   29. within the Logical SFF we assume that Netvirt is creating the transport zones beforehand (edavsua, 17:29:51)
   30. This is for Carbon (edavsua, 17:30:39)
   31. Non-NSH is for Carbon as well (edavsua, 17:31:22)
   32. we need to continue using the old-netvirt (edavsua, 17:32:15)
   33. Ericsson's priority will be the Non-NSH for Carbon (edavsua, 17:33:51)
   34. SFC will be integrated with Genius and using either NSH or VXLAN as transport, but reusing the same code as much as possible (edavsua, 17:34:53)
   35. we'll cancel the meeting of next week (edavsua, 17:38:39)
   36. can we code all the current metadata (NSP and NSI) in the VNI? (edavsua, 17:39:54)
   37. we can use one of the first tables (ingress) to load registers with the VNI to be used later in the pipeline (edavsua, 17:41:07)
   38. we'll have limitations in the number of service functions due to the limited amount of bits (edavsua, 17:42:25)
   39. Eric: how are we going to differentiate the NSH and the Non-NSH in the model? (edavsua, 17:43:59)
   40. we'll have different transport flag (edavsua, 17:44:31)

Action items

1. (none)

People present (lines said)

1. edavsua (36)
2. ebrjohn (5)
3. odl_meetbot (3)
4. shague (1)
5. dgranados (1)