- From: Willy Tarreau <w@1wt.eu>
- Date: Thu, 3 Jul 2014 21:34:35 +0200
- To: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Cc: Daniel Sommermann <dcsommer@fb.com>, Martin Thomson <martin.thomson@gmail.com>, Greg Wilkins <gregw@intalio.com>, HTTP Working Group <ietf-http-wg@w3.org>
On Thu, Jul 03, 2014 at 06:59:57PM +0000, Poul-Henning Kamp wrote: > In message <53B5A00F.9030705@fb.com>, Daniel Sommermann writes: > > >6.9.1 has it: > > > >A sender MUST NOT allow a flow control window to exceed 2^31 - 1 bytes. > >If a sender receives a WINDOW_UPDATE that causes a flow control window > >to exceed this maximum it MUST terminate either the stream or the > >connection, as appropriate. > > I suggest we raise that limit to 2^63-1 bytes. > > 2^32 bytes only take a few seconds seconds at 100 Gbit, and files > larger than 2^32 are common as muck these days. Well, 2^31 should be about 160ms at 100Gbps, but you can only achieve such rates over TCP on very short links because TCP itself will not allow you to grow a window that large, so in practice you'll need some round trips at least for TCP acks. So in practice there will be more than one TCP connection at these rates when observed out of the datacenter, thus I don't think there's any gain in mandating a window larger than 31 or 32 bits. Willy
Received on Thursday, 3 July 2014 19:35:05 UTC