Re: Size of window variables

On Thu, Jul 03, 2014 at 07:47:08PM +0000, Poul-Henning Kamp wrote:
> In message <20140703193435.GB18168@1wt.eu>, Willy Tarreau writes:
> 
> >thus I don't think there's any gain in mandating a window
> >larger than 31 or 32 bits.
> 
> Right now ?  No.
> 
> But I find the margin is too small for HTTP/2, we should design for
> at least 10 years lifetime before people run into trouble.

Given that TCP cannot do better with its 32-bit sequence numbers, and
anything but TCP/IPv4 seems unlikely to replace it anytime soon, I must
say I'm not worried even for the 10 next years (and don't get me wrong,
I too am concerned about long term).

> Also these four bytes don't add wire overhead so there is no
> performance penalty for being on the safe side.

That's perfectly true, the only cost is 8 extra memory bytes per
stream (4 per direction). I'd say I don't really care about them,
it's only that I'm not convinced that there is at least a *chance*
for them to ever be used. It all depends if there are more risks
of seeing bogus implementations or not just for this.

Willy

Received on Thursday, 3 July 2014 20:04:05 UTC