ForCES Working Group, Wednesday, Nov 7th, 2012 (scribed by David Lamparter ) CHAIR: Jamal Hadi Salim (hadi@mojatatu.com) 1) Meeting started with the chair presenting the agenda and pointing details to remote users. Given the tight schedule, the chair imposed a limitation to 5 min per presentation and 1 question max for the audience on a presentation. WG status ========= 1) Document updates since last meeting: LFBLib has started the publication journey. CEHA will follow. 2) Chair posted request for non-charter-work items in mid-Aug Lots of private mails, but not much public; chair asks people to to post on the mailing list instead of sending private mail. * 4 new drafts which are out of charter have been posted. 3 more promised. * 2 earlier drafts revised. * Additionally, there were more requests for presentations than available slots. 3) Chair gave a brief summary of the current charter and emphasized that this meeting is about discussing possible work items outside charter, then decide either to extend the charter or shutdown WG. 4) Chair went over the summary of all the items; what would require a recharter and what would not (refer to the slides) One question posed was whether the IRS northbound interface required multiple CEs controlling a single FE. The suggestion is that there is need to extend the CEHA draft. Joel Halpern(Ericsson) responded that, yes multiple controllers controlling the same agent are needed but it didnt make sense to change the CEHA draft. ForCES Model Extension ====================== presented by Jamal on behalf of Evangelos Haleplidis(Univ of Patras) Evangelos is asking to allow the model to define complex metadata. The need for this arises from his experiences in dealing with OF datapath modelling. Jamal commented that the model was intentionally simple in this aspect and would be fine for hardware implementation but now problematic for software. Ed Crabbe suggested to make metadata opaque instead and cover everything. ForCES Parallelization ====================== presented by Joel Halpern(Ericsson). The ForCES model does not support multiple LFBs performing in parallel. To resolve this, Joel presented a propasal which adds two new LFBs: A splitter LFB and a merger LFB. The splitter allows for modelling of packets sent to several LFBs in parallel and the merger allows for syncing from a parallel set of LFBs to a single point. Bhumip Khasnabish(ZTE) asked if this useful proposal should be considered to be in scope for current charter. Jamal: It is outside. we are on a journey looking for what's needed if we are to re-charter InterFE LFB =========== Presented by Damascene Joachimpillai(Verizon) DJ pointed out this was a problem that they need to resolve to use ForCES currently. They want to pass the packet between FEs. This issue is out of scope of charter. It requires modifying the packet adding new fields to support InterFE LFB metadata as presented. Joel Halpern: This is a problem that i fully support needs to be solved. The packet format could be simpler than presented if metadata is attached to packet redirect TLV. Jamal: This seems mostly based on CE-redirect packet; maybe a redefine in the same way DJ: I need to review and will get back to the WG. XEM interface ============= Presented by Jamal Hadi Salim(Mojatatu Networks) Issue is out of charter scope. Jamal said from experince, it has become obvious that they could model the config plane of an FE as an LFB and pointed people to his presentation in IETF84. He described a use case where he feels standardization would be useful. A single NE, with a physical FE booting up multiple virtual FEs, He wants CEM -> FEM to be told to get instances booted up but in the meantime use the same socket interfaces for all vFEs. This would require a reserved CE/FE to be used for proxying. He believes it is also useful to define a basic standard config for C/FEM. Since this is outside the scope - it would require a recharter. Forces-impl. OF switch + Demo ============================== Presented by Omar Cherkaoui(UQAM) Omar talked about their ForCES implementation on a 100G switch. The purpose was to evaluate flexibility of ForCES implementing OF. The implementation uses an OF switch datapath they implemented earlier on an Ezchip 4 device. The implementation is derived from the OF datapath representation as described in the OF ForCES model draft where Omar is a co-author. Omar pointed people to the demo running in the back of the room where they are welcome to visit afterwards. He also mentioned the demo will be running at Salon E on Thursday for anyone who may not have time after the meeting. The chair cut short Omar's presentation short because of lack of time. Working Group Activities ------------------------ Taken over by chair.. Chair asked the AD (Adrian) how he would like to proceed. Adrian: Why dont you get started then we'll see where to proceed? Jamal: On the model extensions. Is anyone interested in that work? By that i mean will do the work, review or would like to see it pursued because they think it is useful. Bhumip: is IETF going to blindly endorse OpenFlow? Jamal: I defer to the AD. Adrian Farrell: No. Bhumip: So then i dont see the need to be questioning the items presented here today. It is clear to me we need to fill whatever gaps that would complete the work. i.e look what can be delivered and expanded since people are saying they need it. Edward Crabbe (google): It is interesting work but focus seems to be academic and not commercial implementations. Jamal: If by commercial you mean some big vendor has implemented, then the answer is no. The big vendors have certain bussiness models which this would break. Naturally, they will not implement or endorse. Edward: But are there commercial entities using it? Jamal: Sure. Unless small companies dont count as commercial. If you want a big company perspective(not big vendor, but a user) maybe DJ, if ok, can talk to it. DJ: We looked at a few things shipping that would enable us to deliver what we want and at the mechanism that will actually work well in controller-datapath environment and we saw that ForCES was the best fit for us. OpenFlow does not have the constructs we need. If I have a box that runs OF - I will look at it as a single FE. and will worst-case use OF as black box and put ForCES on top Dan Frost (CISCO) Dan: Maybe i can provide big vendors perspective. The work here should not be judged based on whether a big vendor has implemented or not. It is feasible a big vendor doesnt accept something initially, but maybe in a future time people come back and say "we need this". It is clear lots of work was put into ForCES architecture with good ways to extend it. Future work depends on how architecture develops. Ed: ForCEs stuff is relatively a lot more mature Dan: people have different ideas about whether additional work needs to be done and I think we should let that happen. Ed: I agree the parallelism stuff is needed very much needed Jamal: There are other suggestion for work that are not in list, not enough time to allow for presentations. Adrian: Jamal, can you name ther suggestions? Jamal: Well, since Joel is not here to shoot at me: CE-CE is one aspect. But perhaps that could be achieved without any changes to the architecture. I could come up with a full list if needed. Adrian: Even without a WG, one can write drafts & do RFCs Ed agrees with Adrian. Adrian: if not enough critical mass, we cant have a WG; however, if there is a group of people wanting to work together to work on "forces next-gen" thats different. Jamal: hard question to answer here at the meeting. Also - what would be a critical mass? Adrian: how many people are tourists? (show of hands please) hands -> about 15% of the room. Jamal: There just as many who are not present here that have interest in this work. Adrian: starting with small base may not be enough Are there people who want to build and deploy? Jamal: We have people who want to build & deploy You saw DJ's presentation - he wants to deploy, the framework as is doesn't work for him; what he needs is out of charter. Joel's request as well is out of charter. Adrian: I don't care about charter; if everything was allowed, would it still be a WG? Jamal: I think it gets back to definition of "critical mass?" Adrian: We need to gather enough momentum. So lets defer towards collecting that critical mass. question left open for discussion Followup will happen on the list.