Re: SPDY Review

On Sat, 09 Jun 2012 00:55:42 +0200, Mike Belshe <mike@belshe.com> wrote:

>
> OK.   Yes, we know that the dictionary only helps a small amount.  But it
> does help.  The good news is that its basically free to use the  
> dictionary,
> so I don't really see any downsides.
>

I would guess that there are device manufacturers that would object
that 1k data for 100 bytes saving when establishing a connection is
not free nor worthwhile. That is not our main objection though.

The dictionary is a fixed magic number that has to go into every SPDY
implementation. Since the dictionary is not well researched there will
likely be future proposals for replacements. If approved this creates
either the necessity for some handshaking or a library of
dictionaries. Also, no matter how perfect we make it now, thinking of
all different existing user agents and use cases, it will lose its
relevance over time as these change.

I'm basing the assertion that it is not well researched on the fact
that it is easy to propose improvements on it. The most major ones
would be adding tokens like ":method", ":version" and ":scheme" to the
dictionary (they were not updated in the dictionary when the protocol
was changed between version 2 and 3, and gives an instant 2%
improvement) and to split the dictionary for the request context and
the recipient context, as the headers are mostly mutually exclusive.

If you look at the compressed data you will see that 50% of the
characters in the dictionary are used in less than 5% of the cases.
7% are never used, not even accidentally like creating " Apple" from
" Apr" and "Implemented".

Bottom line is that the cost is insignificant, the savings are
insignificant, but the maintenance is an issue.

/Martin Nilsson

-- 
Using Opera's revolutionary email client: http://www.opera.com/mail/

Received on Monday, 11 June 2012 16:27:35 UTC