Greetings, all, I've proposed a lightning talk on this subject for Dubai, but it occurred to me that this working group might be a better place to answer this question: how prevalent is the use of passive latency measurement (e.g., by looking at DNS request/response times, or TCP seq-ack / timestamp delays) in everyday network operations, and for what is it commonly used?
I ask this question in the context of a very specific design proposal before the IETF's QUIC working group. QUIC encrypts almost all of its header information beyond that needed for its crypto handshake, which makes equivalent techniques to those used in TCP impossible: you see no ACK numbers, and no timestamps, so it's impossible to match an upstream QUIC packet with its downstream counterpart. However, a recently proposed addition to the spec (https://github.com/quicwg/base-drafts/pull/609) would add a single bit to the header which would be explicitly intended for passive round trip time measurement. Since this adds a small amount of complexity to the wire image, one useful piece of input to the decision whether or not to add this measurability is how useful it is, in terms of operational impact of having passive measurability go away as TCP flows are replaced by QUIC flows. Thanks, cheers, Brian
signature.asc
Description: Message signed with OpenPGP
