- From: Lucas Pardue <lucaspardue.24.7@gmail.com>
- Date: Sun, 26 Mar 2023 06:54:25 +0100
- To: HTTP Working Group <ietf-http-wg@w3.org>
- Message-ID: <CALGR9obw3HFQY6dqG5f_a5c-EuO1gpEMQ+tB8xVkZqzMxkU+Nw@mail.gmail.com>
Hi folks, Please see the attached notice for a proposed send-order parameter for extensible priorities. Recall that the recommended scheduling guidance [1] is that non-incremental responses of the same urgency be served in stream ID order. This closely maps to typical web cases where a user agent has a good chance of dispatching requests in a sensible order. However, there are some use cases where stream ID order might not be optimal. A late discovered item could be more valuable to receive than others that are being queued up to serve - effectively getting HoL blocked in the sender's scheduler. While it might be possible to work around that by smarter or more elaborate use of urgencies, there are some factors and constraints that make that difficult in practice. An alternative to juggling urgencies is the send-order parameter, with the idea being to enhance the current scheduling guidance. The scheduler would effectively operate over three groups per urgency level in this order: 1) The non-incremental send-order group, served in send order 2) The non-incremental group, served in stream order 3) The incremental group, served in round-robin fashion One example of use case for this parameter is a LIFO style media streaming case (such as those being discussed in the Media over QUIC WG), where the last media items are more important than early ones. If all items have the same urgency, LIFO can be trivially achieved by incrementing the send-order parameter for each one and effectively never giving the scheduler group 2 or 3. Another example use case is explicitly letting an item jump the group 2 queue by providing a send-order of any value. This could potentially be most useful for server-side overrides, whereby it is difficult to exchange connection-local stream IDs in a way that would allow an origin behind the proxy to make good decisions. Instead, they could rely on a fairly simple send-order mapping table to influence behaviours downstream. Cheers Lucas [1] - https://www.rfc-editor.org/rfc/rfc9218.html#name-server-scheduling ---------- Forwarded message --------- From: <internet-drafts@ietf.org> Date: Sun, Mar 26, 2023 at 6:15 AM Subject: New Version Notification for draft-pardue-httpbis-priority-order-00.txt To: Lucas Pardue <lucaspardue.24.7@gmail.com> A new version of I-D, draft-pardue-httpbis-priority-order-00.txt has been successfully submitted by Lucas Pardue and posted to the IETF repository. Name: draft-pardue-httpbis-priority-order Revision: 00 Title: HTTP priority order extension Document date: 2023-03-26 Group: Individual Submission Pages: 8 URL: https://www.ietf.org/archive/id/draft-pardue-httpbis-priority-order-00.txt Status: https://datatracker.ietf.org/doc/draft-pardue-httpbis-priority-order/ Html: https://www.ietf.org/archive/id/draft-pardue-httpbis-priority-order-00.html Htmlized: https://datatracker.ietf.org/doc/html/draft-pardue-httpbis-priority-order Abstract: The send-order parameter for the HTTP extensible prioritization scheme allows explicit ordering indication independent of request order. This can be used to as an additional input signal to scheduling decisions, to support alternative sending behaviors. The IETF Secretariat
Received on Sunday, 26 March 2023 05:54:49 UTC