- From: Amos Jeffries <squid3@treenet.co.nz>
- Date: Sat, 12 Jul 2014 20:51:21 +1200
- To: ietf-http-wg@w3.org
On 12/07/2014 4:07 p.m., David Krauss wrote: > > On 2014–07–12, at 12:01 PM, Roberto Peon <grmocg@gmail.com> wrote: > >> I used to think sending the priority from the server was a decent >> idea, but then I though about exactly what Martin is worried >> about: If the server informs the client about how it is >> prioritizing, and the client disagrees, then it attempts to change >> it. .. and if the server then advertises the same thing... .. rinse >> repeat. >> >> You can get into nasty situations. >> >> It is more robust to simply have the client state the priority if >> it cares. > > How does the client know whether it cares to reprioritize if it > doesn’t know the current priority? > > As I mentioned, setting the initial priority of every push to > default=16 is both nonviable and meaningless. > > What’s left is to attempt retrospective detection based on relative > stream rates, which is slow, difficult, and error-prone. > > A feedback loop is easily averted by not sending more than one > PRIORITY per stream (!). We’re not talking about PRIORITY frames from > the server here, only a prioritization field in PUSH_PROMISE. > Lets converse then: client: HEADERS (s=1, p=999) "I want this. top-priority!" server: PUSH (s1 + s=2, p=4) "s=1 comes with extras. I will send in background, real SLOW" client: PRIORITY (s=2, p=999) "I want those top-priority as well!" Seems reasonable information exchange. Still no obligation on the server to actually use the client hints. Amos
Received on Saturday, 12 July 2014 08:51:55 UTC