Routing Reflections @ IETF 96: A Routing AD's Perspective
Jul 26, 2016
While I always go to IETFs with excitement to participate in the discussions, see the many friends and colleagues there, and hear about new ideas, this particular IETF was special since Juniper was the Host and I'd been active on the hosting committee. The t-shirts and fans were welcomed; even though it wasn't absurdly hot this time, the ability to adjust temperature inside crowded meeting rooms was appreciated. The Social was elegant, tasty, fun, and filled with good discussion. Kevin Walker'stalkon advancing security using next-generation protocols went well.
On to the more exciting things!
With my encouragement and focus, there is more engagement between the Open Source communities and the IETF. Of course theIETF Hackathon had more participants than ever and we experimented with having a place for it in the IETF Lounge all week. More exciting to me is that theBabel Working Group met for the first time and it was filled with enthusiasm, focus, and good discussion. Juliusz Chroboczek suggested a middle ground that didn't immediately throw away backwards compatibility but offers the freedom for necessary improvements. Most of us really enjoy technology, so seeing Dave Taht'spresentationabout Babel on hacker-boards was fun. Of course, having folks participating eagerly from Apple, Google, AT&T and others is even more encouraging. In RTGWG, Sam Aldrin from Google presented adraftdefining the gRPC protocol; the use-casesincluded telemetry similar to Juniper's JVision. Benoit Claise, the Management AD, will be guiding this work. I and my fellow Routing ADs also had an open side-meeting where we invited those participating in Open Source related to the Routing area to come talk to us and each other about how interactions were going. We know have a plan for a mailing list (email@example.com) and awiki and will meet again in a less impromptu fashion next IETF to talk more.
The work on YANG models in the Routing Area continues; the amount of detail and the fact that implementations really do differ greatly in the abstractions applied makes this very time-consuming and challenging work. Of course, that's just a sign that the benefit of having a lot of common configuration and operational data easily accessible across many vendors is high. We're trying to do it once so that all the operators don't have to - over and over and over and over again. With Benoit Claise, the Routing ADs (Alvaro, Deborah and I) are pushing to try and have a coherent set of routing YANG models done in the next year. This is based on dependencies between them, of course, and also implementations to make certain that they connect well. There has been some issues around how intended state is represented in models;NetModhas made a decision and it should now be possible (with guidance that we are working on) to move ahead with these.
OAM and telemetry were another focus. I hadchartered an Overlay OAM design teamto look at what's possible for defining OAM for the different encapsulations coming from the NVO3 (GUE,VXLAN-GPE,GENEVE), SFC (NSH) and BIER working groups. There are three basic aspects to be considered. First is providing guidance for how many bits to allocate in a header for OAM and how they should be used. Second is considering a common OAM format and data that can be reused by multiple encapsulations. Third is examining cross-layer correlations and facilitating trouble-shooting where the number and ordering of layers isn't as fixed as traditionally assumed;layer-transcending tracerouteis an example approach. At this IETF, the design team finished and their mailing list (firstname.lastname@example.org) is now open for a broader description of how this work should proceed. Telemetry keeps coming up as a point of interest - whether that be using gRPC to send it ordataplane probesas discussed at IETF 95.
InISIS, adraftgave a different approach to handling the spine-leaf topology. This is of personal technical interest to me since I co-authored adraft suggesting using virtual multi-instances to solve this and related problems. Shraddha Hegde's approach works for both OSPF and IS-IS and requires no protocol changes - which is good when a network has equipment where code changes are hard. It's good to see solutions emerging to handle the same set of problems, since that indicates others are seeing the issues as well.
In OSPF, we had a good discussion about encouraging more document reviews. Of course, in the IETF, a person sending a review also gives an indirect signal of how useful or relevant the draft is. That's why the Internet Draft Review Teams, that Nalini Elkins on the IETF Mentoring team is facilitating, specifically have the participants figure out what drafts they are interested in reviewing instead of requesting suggestions from the WG Chairs.
I am pushing NVO3 towards a conclusion around the three different encapsulations (VXLAN-GPE, GENEVE, and GUE) that they've been considering. At the face-to-face meeting, there were technical objections raised to each of them. On the mailing list, the discussion continues. It's a challenge when there are entrenched technologies widely deployed; we need to look ahead to see that a useful impact on the industry can still occur even if it may be a few years out. The NVO3 working group hasn't really started seriously discussing control planes yet; it's not yet clear if the same difficulties would occur if NVO3 does start that discussion.
In RTGWG, the discussion around how to handle IPv6 PA multi-homed networks (homes, small enterprises, etc) as requested in IETF 94 by v6ops continued. Jen Linkova presented adraftthat intends to describe the full problem space. Thevirtual interim really helped forward this conversation.
I could go on for even longer about interested work being done inBESS,I2RS,IDR,BIER(thinking about a second encapsulation), and more - but you should go look for yourselves!