Re: RT/NT support for HTTP/2 priorities/weights?

On Thu, Sep 4, 2014 at 5:19 AM, Peter Lepeska <bizzbyster@gmail.com> wrote:

> Yes. Dependencies/weights in RT feedback can tell us which objects on the
> page had the biggest impact on page load time. These are the objects we
> most need to include as resource hints. In fact, based on dependency/weight
> information, a hinting service, which I'm defining as a module which takes
> in RT feedback and generates resource hints, may decide not to include all
> objects on the page as resource hints b/c the benefit of speculatively
> preloading certain objects, based on dependencies/weights, does not justify
> the cost.
>

> Beyond this, one can imagine a caching algorithm favoring blocking objects
> (exclusive=true) over non-blocking objects in its cache replacement policy
> algorithm.
>
Lastly, to the extent that we will be able to express priority in resource
> hints, which I argue for here (
> http://lists.w3.org/Archives/Public/public-whatwg-archive/2014Aug/0216.html),
> a great source for the priority information is RT feedback.
>
> Do those use cases make sense?
>

With HTTP/1 the priorities are hard-coded and determine the request
dispatch order, and in HTTP/2 the priorities are already explicitly
communicated to the server within the protocol itself. It seems like you'd
have all the necessary data already? E.g. look at priority of incoming
stream, use that as a signal on the server to inject an appropriate hint
for a subsequent visitor, etc.

It doesn't seem that exposing this data on the client offers much
additional benefit?

ig


>
>
> On Wed, Sep 3, 2014 at 6:57 PM, Ilya Grigorik <igrigorik@google.com>
> wrote:
>
>>
>> On Sun, Aug 31, 2014 at 8:10 AM, <bizzbyster@gmail.com> wrote:
>>
>>> It would also be great to have this information in Resource Timing API
>>> feedback.
>>>
>>
>> Peter, can you elaborate? I'm not sure what you're asking for here
>> exactly... Exposing browser-set priority levels? What's the use case?
>>
>> ig
>>
>
>