- From: Zhong Yu <zhong.j.yu@gmail.com>
- Date: Fri, 21 Feb 2014 16:37:46 -0600
- To: Martin Thomson <martin.thomson@gmail.com>
- Cc: Roberto Peon <grmocg@gmail.com>, Mark Nottingham <mnot@mnot.net>, HTTP Working Group <ietf-http-wg@w3.org>, Patrick McManus <mcmanus@ducksong.com>
On Fri, Feb 21, 2014 at 3:03 PM, Martin Thomson <martin.thomson@gmail.com> wrote: > On 21 February 2014 12:46, Zhong Yu <zhong.j.yu@gmail.com> wrote: >> I also agree with Bjoern's point that >> gzip/deflate is pretty slow; enabling it may actually decrease >> throughput in a lot of deployments. > > I was under the impression that it was compression speed that was the > concern. Here we're talking about decompression only. A server is > under no obligation to use compression; if the server wants more speed > and thinks that avoiding the compression schemes on offer is the best > way to achieve that, then go for it. > > The question is what options we require clients to support so that > servers always have the option to select something better than > "identity". Yes. And the argument is that if a server can do gzip, it can do deflate as easily, so there's no point to give it both options. (Compression of requests is probably out-of-scope here, but since http/2 server implementation is a heavy task requiring abundant resources to accomplish, mandating servers to accept compressed requests may not be too much to ask? Or is it the last straw that breaks the camel's back?) Zhong Yu
Received on Friday, 21 February 2014 22:38:13 UTC