Re: expected substantial and measurable improvements in WG charter

HTTP/2 substantially and measurably improves end-user perceived latency.
It doesn't in all cases, however, when the SHOULD about using one
connection is followed, as it should be (not following this will eventually
make things worse).

The most important of these is when the transport makes having 6
connections outperform one link with the same peers.
Fixing this (the transport) is *the* thing to do.
For congestion algorithms like cubic you need to have appropriate initcwnd
settings (between 10 and 32), and you need to react appropriately to loss
(e.g. as having K (or 6) connections would).
If you're using different congestion avoidance algorithms, the tweaks to
make the transport behave sanely are likely to be different in the details,
and the same in principle.

We've said this a number of times, so I don't know why there is any
impression that there is something proprietary here.


On Wed, Mar 11, 2015 at 6:30 PM, Matthew Kerwin <>

> On 12 March 2015 at 00:10, Yoav Nir <> wrote:
>> Give it a year, and we’ll read about people outside the small group
>> attempting to deploy HTTP/2. Maybe they’ll love the improved user
>> experience. Maybe they’ll measure it and conclude that it didn’t help and
>> didn’t hurt. Maybe they’ll find that it makes things worse and revert. Then
>> we might be able to make a better when and why document.
> Is that an argument for making HTTP/2 an experimental standard? (I'm not
> whipping a dead horse, but genuinely questioning the meaning of
> "experimental" vs "proposed" standards.)
> Is it well and truly time for us, as well, to start working on the
> apocryphal "how to use HTTP/2" document that's been mentioned from time to
> time over the years? I'd start, but I don't know where to begin (since I'm
> partly the target audience, wanting to know why I need to use TLS to work
> with Chrome and Firefox, etc.).
> --
>   Matthew Kerwin

Received on Thursday, 12 March 2015 01:50:56 UTC