#opnfv-sfc: opnfv-sfc-weekly

Meeting summary

1. 1. Daniel Smith (lmcdasm, 14:06:24)
   2. tbachman (tbachman, 14:06:27)
   
   
2. Agenda (tbachman, 14:06:42)
   1. Nicolas Bouthors (bouthors, 14:06:45)
   2. Dave Dolson (DaveD_, 14:06:49)
   3. Paul Quinn (paulq, 14:07:08)
   4. https://wiki.opnfv.org/service_function_chaining (ebrjohn, 14:07:33)
   5. ebrjohn says we’re skipping the clustering action item for now (tbachman, 14:08:44)
   6. paulq asks why opnfv doesn’t use specific mailing lists (tbachman, 14:09:40)
   7. ebrjohn says that opnfv early on decided to have a single list and use filters (tbachman, 14:10:06)
   8. bryan_att says that you can set filters using mailman (tbachman, 14:10:52)
   9. ebrjohn says the agenda is carrying on with the slides that cover initializing OPNFV and SFC, and openstack orchestration (tbachman, 14:11:32)
   
   
3. OPNFV SFC initialization (tbachman, 14:11:51)
   1. https://docs.google.com/presentation/d/1gbhAnrTYbLCrNMhMXin0lxjyg-7IHNPjrlBTIjwAzys/edit?usp=sharing Slides for OPNFV/SFC (tbachman, 14:13:00)
   2. ebrjohn says he added the GBP endpoints, added SFC/OpenStack coordinator (tbachman, 14:14:10)
   3. paulq says the GBP depiction might be a bit off — wrt to the classifier; he’ll collect his feedback and provide it in an email to ebrjohn (tbachman, 14:14:42)
   4. bryan_att asks if the coordinator is outside ODL, or something else (tbachman, 14:15:34)
   5. tbachman says oops (paulq, 14:15:48)
   6. ebrjohn says there’s a configuration hole: SFC needs to know information about the VMs (e.g. addressing); this is one of the functions of the coordinator (tbachman, 14:16:32)
   7. lmcdasm notes that most of the info is available from the neutron DB (tbachman, 14:18:05)
   8. lmcdasm says if you want to know the port definition, then you can query neutron for this information; one of the catch-22’s is if the VNF or Service Function is set up as a VM, then neutron may not be aware; otherwise, this is a simple query to neutron (tbachman, 14:20:13)
   9. paulq asks if you can find the relationship between a VM (e.g. a firewall) and the OVS instance it’s connected to (tbachman, 14:20:29)
   10. lmcdasm says nova fires up the VM, neutron attaches the port; openstack isn’t aware of what the VM is doing (e.g. firewall); but you can associate the port with the vritual NIC for that VM (tbachman, 14:21:03)
   11. lmcdasm says you can’t have 2 ports on the same subnet, so you know which port is connected to a given subnet (tbachman, 14:21:48)
   12. rapenno says from an ODL provisioning perspective, you create a Service Function, and then goes to an orchestrator to request a VM to get the information (tbachman, 14:22:19)
   13. bryan_att says he’d like to hear more information on how the life cycle works (tbachman, 14:23:20)
   14. paulq says he’d like to get a bullet-point list and API pointers so we can see how this looks and works (tbachman, 14:23:44)
   15. lmcdasm says his initial understanding was that it was a pull model from openstack; this new model suggests that ODL is interacting with OpenStack to create VMs for things (e.g. firewalls) (tbachman, 14:24:58)
   16. ebrjohn says we should create two mini-groups: 1 to work on the coordinator, another to work on a setup (tbachman, 14:25:36)
   17. ebrjohn says there are two use cases: reactive approach and proactive approach; the first is reacting to VM creation, the second creates VMs (tbachman, 14:26:08)
   18. lmcdasm is in favor of picking one approach or the other, rather than both (tbachman, 14:26:27)
   19. ebrjohn says we need work on both the coordinator, and work on the piece that brings everything up (tbachman, 14:28:52)
   20. lmcdasm says that we have everything with the exception of the coordinator today (tbachman, 14:29:14)
   21. paulq asks what is a service function and what is it’s relation to OVS (tbachman, 14:29:59)
   22. lmcdasm says a service function is something an application does; the relation between an SF and OVS is that OVS is the fabric that the SF runs on (tbachman, 14:30:30)
   23. ebrjohn says we need a mapping for a service function type (e.g. FW) to actual glance images (tbachman, 14:30:57)
   24. bryan_att says we should take that as part of the coordinator discussion — he’d like to address this at the upcoming hackfest/summit (tbachman, 14:31:25)
   25. bryan_att says the reactive and proactive approaches should be part of the bullet-list analysis (tbachman, 14:32:45)
   26. lmcdasm says something missing is what do you want to run with those applications (tbachman, 14:32:57)
   27. lmcdasm says we should start simple — pick to things that you’d like to chain today and see how to make that work (tbachman, 14:33:59)
   28. paulq says firewall and DPI would be a good first start (tbachman, 14:34:08)
   29. bryan_att says as noted VNFD is the place where the metadata about “what type of function this VM provides” would live (tbachman, 14:34:29)
   30. lmcdasm says the first thing a mini-group should do is define the specifics — what is the VM, what is the glance image, etc. (tbachman, 14:35:28)
   31. bryan_att says the purpose of each vNIC on a VM also needs to be clear in the metadata, so that the proper port can be chained, e.g. "Service", "admin", "storage", etc (tbachman, 14:36:21)
   32. DaveD_ and bouthors are interested in working with the architecture mini-group (tbachman, 14:38:16)
   33. ebrjohn wants a mini group to investigate how to stand up architecture shown in his “OPNFV SFC Initial NW Topology, V2” slide (tbachman, 14:39:08)
   34. bryan_att notes that what the user wants and what the service provider wants might be two different perspectives — we should take this into account when looking at the architectures (tbachman, 14:40:34)
   35. lmcdasm says he wasn’t clear from ebrjohn’s email whether this includes enabling/disabling or configuring ports on physical switches (tbachman, 14:42:45)
   36. ebrjohn says he wouldn’t want to remove any functionality we already have — we should just be adding to it (tbachman, 14:43:09)
   37. lmcdasm says an end-user asks for creation of a service function, but the service provider might have to provision the virtual and physical resources for this; do we want to supercede this and have SFC be in control of this? (tbachman, 14:43:54)
   38. repenno agrees — we have to decide who is the master of a give vSwitch (tbachman, 14:44:18)
   39. lmcdasm says we already have an injector that reads an OVF file and deploys it into openstack (tbachman, 14:49:01)
   40. ACTION: repenno to create a workflow diagram for SFC and will email to the OPNFV-SFC group (tbachman, 14:50:56)
   41. https://docs.google.com/document/d/1CQ6HODjk-sa5jSdZm9l6ygxc3TWTE8Zpgq3_NFbK1mc/edit?usp=sharing Document that shows the SFC workflow (tbachman, 14:51:26)
   42. bryan_att asks to put links in OPNFV-SFC wiki to any demos, code, etc. that can be shared about what’s been done before (tbachman, 14:51:54)
   43. https://docs.google.com/document/d/1CQ6HODjk-sa5jSdZm9l6ygxc3TWTE8Zpgq3_NFbK1mc/edit#heading=h.rrqnx1heyzli Agenda for upcoming Hackfest (tbachman, 14:53:51)
   44. lmcdasm says we should bring in icbts (Jamie Goodyear) or others who are karaf experts (tbachman, 14:53:59)
   45. DaveD_ asks what the forum is for discussing slide 5 (tbachman, 14:54:26)
   46. ebrjohn says the best place for now is the mailing list (use tag/filter in brackets) (tbachman, 14:54:46)

Action items

1. repenno to create a workflow diagram for SFC and will email to the OPNFV-SFC group

People present (lines said)

1. tbachman (89)
2. collabot (9)
3. bryan_att (6)
4. DaveD_ (5)
5. ebrjohn (4)
6. paulq (4)
7. bouthors (2)
8. lmcdasm (2)
9. MR_Sandvine (1)