Re: Header Table and Static Table Indicies Switched

On 2 August 2014 11:02, Jeff Pinner <jpinner@twitter.com> wrote:

> Looking at the annotated headers for a single domain, on each request
> there are:
>
> 16 unique header names annotating authentication information
> 5 unique header names annotating rate-limiting information
>
> Of those headers, all but 1 will take 2 bytes to encode, increasing
> the size needed to reference just these names alone from 21 bytes /
> request to 41 bytes / request.
>


Jeff,

that is just not correct.

You get 65 entries in the header table that have 1 byte to encode.
That's quiet a lot of entries so before you can complain about 2 byte
indexes, you have to demonstrate how you will get to 65 entries  (without
using a dumb encoder that indexes fields that are used only once).

The evidence from the test data publicly available is that most connections
do not get to 65 entries in the header table.  Very few fields needed 2
byte lookups.

Perhaps there is an argument for being able to send an indexed literal name
with a null value, so we can add just the name to the header table for use
with custom fields whose value changes every request - but that is an
optimisation that equally applies regardless if static indexes are high or
low.

regards












-- 
Greg Wilkins <gregw@intalio.com>
http://eclipse.org/jetty HTTP, SPDY, Websocket server and client that scales
http://www.webtide.com  advice and support for jetty and cometd.