- From: Willy Tarreau <w@1wt.eu>
- Date: Sat, 31 Mar 2012 09:13:33 +0200
- To: Roberto Peon <grmocg@gmail.com>
- Cc: Mark Watson <watsonm@netflix.com>, Mike Belshe <mike@belshe.com>, "William Chan (?????????)" <willchan@chromium.org>, "<ietf-http-wg@w3.org>" <ietf-http-wg@w3.org>
Hi Roberto, On Sat, Mar 31, 2012 at 03:22:20AM +0200, Roberto Peon wrote: > HTTP is widely deployed, but most certainly not well understood :/ > > In any case... a number of clients and servers have implemented SPDY, and > haven't had any problems with deployment, nor big problems with > implementation. > > What I understand that you're proposing is to build off HTTP/1.1's current > framing, but bolting on multiplexing (via response IDs), interleaving, and > prioritization. Note that you'll have to also do interleaving on the > request path to avoid HOL blocking there, e.g. when someone decides to POST > a video. I don't believe that this is trivial to make backwards > compatible... in fact, I think that the assumption that "building off of > current HTTP framing semantics is easier" is just plain not true. Hell, one > of the reasons I started SPDY with Mike was my more-than-slightly-strong > distaste for HTTP's framing after having to write a production-quality > framer for an intermediary that sees a lot of traffic (from billions of > people). It is the details that get you-- those niggly little pesky > annoying things. > > SPDY doesn't have that problem with framing. The frames are well defined > and you don't have to dip into the application layer at any time to figure > out how to frame the dang frames. I hope that HTTP/2 is similarly easy to > implement with few (if any) corner cases for the framing. > > Framing is just the wrong place to spend complexity... it makes far more > sense to spend it on features that improve performance, latency, or > security, at least in my opinion. +1 Willy
Received on Saturday, 31 March 2012 07:14:01 UTC