> There was a discussion in the NVO3 WG meeting in Berlin following strong > advice from our Area Director that we need to come to a consensus on > converging on a common encapsulation. Two sets of questions were asked: > (1) Should the WG move forward with one standards track encap? > (2) For a given encap, do you have significant technical objections? >
I prefer to see (1) the WG moving forward with one standards track encapsulation. I think this is the best way for NVo3 to be productive for the majority of end users, as opposed to whoever finds NVo3 useful at the moment (??). *OFF TOPIC* please fork the thread if anyone is inclined to reply: An outsider's observation as someone who stopped following NVo3 a couple of years back because it didn't seem to be making progress: There is an embarrassing amount in common between the three drafts, maybe more than half of each proposal could essentially be identical. The sticking points perhaps should be pulled out for special consideration in their own threads, which so far from what I've read seem to be: 1) Extensibility: should it be done? How to do it? With a shim such as NSH? With reserved flags and a possible version increment when they are used, that can be determined at a future date. Breaking backwards compatibility. With fixed-size option headers that are indicated by flags. But with a header_len field to allow backwards compatibility by simply ignoring all options. With TLV variable length option headers. But with a header_len field to allow backwards compatibility by simply ignoring all options. 2) VNI - should it be included, or an option header like GUE? 3) Security of VNI, is it needed, how to do it? 4) Re-usability of existing H/W blocks with minimal modification (in the case of VXLAN)? Is it really important? I suspect it's not, because frankly the cost of having to implement three new encapsulations where one has some re-usability is much greater than the cost of having to implement one new encapsulation that has no re-usability. 5) Ease of processing in H/W in the case of GUE/GENEVE with their more complex extensibility mechanisms. I am not qualified to hold an opinion here, but it seems like VXLAN-GPE is easiest, GUE and GENEVE similar with GUE probably being slightly easier. Cheers Kris
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
