- From: Patrick McManus <pmcmanus@mozilla.com>
- Date: Thu, 26 Jan 2012 09:32:05 -0500
- To: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Cc: Willy Tarreau <w@1wt.eu>, ietf-http-wg@w3.org
On Thu, 2012-01-26 at 14:19 +0000, Poul-Henning Kamp wrote: > For instance, starting to put "congestion responsiveness", whatever > it is, on top of TCP sounds just plain bogus to me. > > It shoulds like an attempt to work around a problem which should > be solved where it lives (TCP & deep black buffers ?) rather than > inefficiently band-aided around at a random higher different level. > (See also: OSI protocols) > all I'm saying is that spdy's propensity to use 1 connection where HTTP is currently using dozens has significant implications on how congestion control impacts the transport. Most of those implications are very good[1]! A few might not be so great[2]. It seems pretty clear to me that the IETF is the right place to have the discussions about the tradeoffs of how an application protocol interacts with the network. -Patrick [1] probably include reduced bufferbloat, less use of constants like IW and default RTO, better ability to respond to various fast-recovery loss signals [2] due to in-order guarantees of tcp, loss events create fate sharing of that delay between the multiplexed transactions on a spdy connection - something you don't see with HTTP/1 parallel tcp connections.
Received on Thursday, 26 January 2012 14:32:50 UTC