- From: Martin Thomson <mt@lowentropy.net>
- Date: Wed, 03 Feb 2021 14:00:24 +1100
- To: "Ben Schwartz" <bemasc@google.com>
- Cc: "HTTP Working Group" <ietf-http-wg@w3.org>
On Wed, Feb 3, 2021, at 06:37, Ben Schwartz wrote: > The DNS use case is clear enough to me. For telemetry, the utility > seems much lower. Telemetry is (by definition?) not user-visible, so > it's typically not latency-sensitive. Spending one or two RTTs for a > standard TLS connection will not substantially impair the utility of > the telemetry. So telemetry does not directly benefit from the latency gains here, but it does benefit from the reduced cryptographic overheads. Those are non-trivial. > Telemetry also typically doesn't require a reply from the server. It absolutely does. Clients need to know that it was received and accepted. I know that you might tolerate a little loss here and there, but when you drop basic reliability feedback like this you lose more than I think we'd be willing to tolerate. You introduce bias of a very particular type too.
Received on Wednesday, 3 February 2021 03:00:57 UTC