- From: Mark Nottingham <mnot@mnot.net>
- Date: Sun, 15 Jul 2012 10:21:18 +1000
- To: HTTP Working Group <ietf-http-wg@w3.org>
Begin forwarded message: > From: Jeff Pinner <jpinner@twitter.com> > Subject: HTTP2 Expression of Interest > Date: 14 July 2012 11:35:06 AM AEST > To: mnot@mnot.net > > Hi Mark, > > Below is Twitter's response to the HTTP/2.0 Expression of Interest. > Could you please forward it to the HTTPbis working group list. > > Sincerely, > Jeff Pinner > Twitter, Inc. > > > This is Twitter's response to the call for expressions of interest: > http://trac.tools.ietf.org/wg/httpbis/trac/wiki/Http2CfI > > 1. Introduction > > Twitter brings people closer to the things they care about. > > Twitter has implemented, deployed, and open sourced both the SPDY/v2 > and SPDY/v3 protocols and is interested in sharing our experience as > well as participating in the development and eventual standardization > of HTTP/2.0. > > 2. Criteria for HTTP/2.0 > > 2.1 Multiplexing > > In order to improve the performance of HTTP, we believe that > multiplexing should be incorporated into the HTTP/2.0 specification. > The performance gains achievable by the reduction in connection > overhead, the pipelining of requests, and the ability to deliver > responses out of order address many of the limitations of the HTTP/1.1 > specification. Multiplexing also helps alleviate the need for complex > performance optimizations that attempt to work around these > limitations. > > 2.2 Encryption > > Twitter strongly believes that the HTTP/2.0 specification should > mandate transport layer encryption. The exchange of personal > information across the web is only expected to increase, and the > privacy and security afforded by encryption at the transport layer > outweighs any difficulties in the implementation. > > Twitter aims to secure all communication with its users, and is doing > so using commodity hardware and openly-available software. We believe > that the various reasons provided for not requiring protocol > encryption are poor security practice, and that the necessary > improvements to TLS and the CA model should not prevent encryption > from being mandated in HTTP/2.0. Requiring encryption in the protocol > is a step that protects against insecure design via omission or > oversight. > > 3. Assessment of the SPDY Protocol > > 3.1 Multiplexing > > SPDY provides multiplexing support that allows for both request > pipelining and out-of-order response delivery. Coupled with request > priority and per-stream flow control, this provides an effective > mechanism to overcome many performance limitations of the HTTP/1.1 > specification. > > 3.2 Compression > > SPDY mandates the use of compression for the HTTP headers. We have > found significant performance improvements due to the resulting > reduction in message size and recommend that compression be mandated > in the HTTP/2.0 specification. However, the additional benefit > achieved through the use of the compression dictionary appears to only > affect the initial message and thus does not seem to outweigh the > additional complexity. We recommend that its adoption be more > carefully considered. > > 3.3 Versioning > > The versioning scheme in the SPDY protocol is incomplete and is > redundant when using NPN to negotiate the protocol. > > 3.4 TLS > > The SPDY protocol does not mandate the use of TLS, however most > implementations negotiate the protocol using the NPN TLS extension. We > recommend that the use of TLS be mandated and that the protocol be > made compatible with the SNI TLS extension. > > 4. Summary > > Twitter recommends that the SPDY protocol be used as the basis for > further work and eventual standardization as the HTTP/2.0 > specification with the additional requirement that transport layer > encryption be mandated. -- Mark Nottingham http://www.mnot.net/
Received on Sunday, 15 July 2012 00:21:46 UTC