On Fri, Aug 9, 2019 at 12:40 PM Kari Hurtta <hurtta-ietf@elmme-mailer.org> wrote: > Kari Hurtta <hurtta-ietf@elmme-mailer.org>: (Fri Aug 9 14:14:06 2019) > > Also new type codes for HEADERS and PRIORITY frames need similar > negotiation > > (because accounting of frames and HPACK is affected; they can not just > > ignored as unknwon frame type codes are ignored). > > Or actually only HPACK is affected. > > ( Only DATA frames are flow control controlled. ) > > If only new PRIORITY frame (type code) is introduced, but no changes on > HEADER frame > (or no new HEADER frame (type code)), then negation is not needed before > sending frame. > > / Kari Hurtta > It really depends. The absence of priority fields in HEADERS *is* priority information for HTTP/2's default priority scheme. Negotiation provides feedback. Without it, a client that opportunistically sends new_priority_scheme instructions and neglects to send the default priority scheme's instructions may cause unintended behaviour when the server ignores the new signal and instead prioritizes everything to the root with the same weight. That results in a round-robin priority scheme which is demonstrably worse for web browsing performance.
Received on Friday, 9 August 2019 12:05:56 UTC