Re: Implementation Notes on Server Push

I think this makes sense. Alternatively we could just add it as 4 bytes to
the (bi-directional) SYN_STREAM.

Even at its current 16 byte overhead using window_update its not a big deal
afaict, though separate settings for things that are logically quite
separate is more efficient and a good idea. I mostly mentioned it because I
was impressed that I didn't find a server unhappy with using a
window_update to grow the window to a value larger than default established
by SETTINGS.

Try doing the rough equivalent to TCP with setsockopt and SO_RECVBUF and
you will get decidedly inconsistent results.




On Wed, May 15, 2013 at 4:36 PM, Martin Thomson <martin.thomson@gmail.com>wrote:

> On 14 May 2013 18:24, James M Snell <jasnell@gmail.com> wrote:
> > "Splitting the initial window" seems wrong to me. Of course, I don't
> > have a better suggestion at the moment, but if I did I'm not certain
> > it would be this ;-)
>
> Sorry, I wasn't clear enough.  That would be crazy-talk.  I meant to
> split the initial window size configuration items.  Rather than "my
> receive window starts at X", it will be "my receive window for
> odd-numbered streams is X but for even-numbered stream it is Y."
>