#opendaylight-meeting: of app coexistences

Meeting started by phrobb at 14:00:58 UTC (full logs).

Meeting summary

  1. agenda bashing (colindixon, 14:01:36)
    1. https://meetings.opendaylight.org/opendaylight-meeting/2015/of_app_coexistences/opendaylight-meeting-of_app_coexistences.2015-09-18-14.05.html meeting minutes from last time (colindixon, 14:05:41)
    2. it sounds like we have two agenda items: Yi presenting and colindixon to provide a summary (colindixon, 14:06:19)

  2. summary (colindixon, 14:06:22)
    1. https://lists.opendaylight.org/pipermail/discuss/attachments/20150910/11fe33b7/attachment-0001.pdf colindixon’s diagram (colindixon, 14:07:40)
    2. alagalah notes that he wants to make two key statements: (1) this about interactions, and (2) he wants to make sure we don’t bake assumptions about where and when things branch (colindixon, 14:14:36)
    3. dlenrow notes that you can totally screw over downstream people (colindixon, 14:14:52)
    4. https://lists.opendaylight.org/pipermail/discuss/2015-September/005689.html colindixon’s stab at defining API (colindixon, 14:15:35)
    5. dlenrow, alagalah, and dlenrow talk about what’s likely to come next and have somewhat differnt views, but that’s something for the future (colindixon, 14:20:43)
    6. there is agreement that if we don’t get to the point where we have composable apps, we will fail (colindixon, 14:23:20)

  3. what’s next (colindixon, 14:23:43)
    1. Yi says he thinks all three proposals could be implmented with the same framework (colindixon, 14:25:01)
    2. further he says he’s implemented a first take at that (colindixon, 14:25:21)
    3. he says he’s already done some work to move GBP and SFC to using this and showing doing a demo (colindixon, 14:26:12)
    4. it’s a new flow programmer API, which includes something new called “write flows” which I think takes the role of partitioning table numbers (colindixon, 14:27:23)
    5. the API provides two key functin calls: addFlowSpace() which adds a rule into table 0 that can direct people to different logical pipeline pieces, and then writeFlow() which then writes into a virtual table space for a given project (colindixon, 14:30:38)
    6. https://git.opendaylight.org/gerrit/#/c/26593/ I think this is the gerrit (colindixon, 14:32:30)
    7. there is some back and forth, colindixon and alagalah ask how much of this is baked into the flow programmer APIs (colindixon, 14:39:50)
    8. for example, we need to be able ot dynamically add new pipelines and allocate tables at least when we first run the controller without having to recompile code, and maybe also actually during runtime (colindixon, 14:40:37)
    9. afredette makes an analogy to running programs on a processor where pipeline snippets are basically functions (colindixon, 14:41:09)
    10. what we’ve talked about so far creates a concept of functions and keeps them from directly interefering with each other, we still need a way to combine them into a program that is then compiled onto a switch (colindixon, 14:43:56)
    11. https://lists.opendaylight.org/pipermail/discuss/2015-September/005689.html colindixon notes that he *tried* to call out a stab at the way to consturct those programs with these four APIs here (colindixon, 14:45:59)
    12. afredette thinks about things pretty concretely with function calls, e.g., where you you explicitly invoke another “function” and pass data in the packet possibly plust metadata (colindixon, 14:47:07)
    13. alagalah asks what the proposal is for handing packets between the logical pipelines (colindixon, 14:48:38)
    14. ACTION: Yi will post the code also for GBP and SFC (colindixon, 14:49:02)
    15. it sound like everyone agrees we want to get some thing done in Beryllium (colindixon, 14:55:35)
    16. afredette and alagalah say that their goal for the first step would be to get OVSDB netvirt and GBP to both coexist with SFC using the same mechanism, that would be a first step (colindixon, 14:56:54)
    17. the next step would probably be to get a small nubmer of people in a meeting to write Java interfaces (colindixon, 14:57:13)
    18. it sounds like afredette, alagalah, Yi, shague, and colindixon would be a good first stab (colindixon, 14:58:16)
    19. alagalah notes that times are hard when Yi is in shanghai and others are east coast (colindixon, 14:58:40)
    20. Uri says that 7a pacific seems to be a good time (colindixon, 14:59:04)
    21. alagalah says, in the meantime, he wants to see all of Yi’s code including patches to SFC and GBP not just the OpenFlow plugin (colindixon, 15:00:23)
    22. getting vishnoianil involved here as well would be good (colindixon, 15:01:09)
    23. ACTION: alagalah will try to find a meeting time (colindixon, 15:01:50)


Meeting ended at 17:00:27 UTC (full logs).

Action items

  1. Yi will post the code also for GBP and SFC
  2. alagalah will try to find a meeting time


Action items, by person

  1. alagalah
    1. alagalah will try to find a meeting time


People present (lines said)

  1. colindixon (48)
  2. phrobb (9)
  3. odl_meetbot (7)
  4. alagalah (3)
  5. abhijitkumbhare (3)
  6. afredette (1)


Generated by MeetBot 0.1.4.