Re: Header Compression Implementation Feedback

On Tue, Jul 9, 2013 at 8:17 AM, Martin Thomson <martin.thomson@gmail.com> wrote:
> On 9 July 2013 07:28, Michael Sweet <msweet@apple.com> wrote:
>> If instead we use a per-stream header table then we avoid this complexity but end up with less effective compression, particularly if clients and servers do not reuse streams for multiple requests.
>
> Perhaps there is a misapprehension about how this works.  The draft
> says: "An HTTP request/response exchange fully consumes a single
> stream."
>
> That means that:
>  1. You only get one set of headers on every stream in each direction,
> with the exception of push promises, of which there can be any number
> on the same stream as a response.
>  2. As a result, unless you have lots of push promises, per-stream
> header compression will net almost zero compression gain.
>

Well, header compression is at the framing layer, which could have any
number of HEADERS frames and PUSH_PROMISE frames sent, but the point
stands... per-stream header tables would be a very bad idea... (in my
opinion it's bad enough that we need to have one for each direction).

Received on Tuesday, 9 July 2013 15:51:06 UTC