Compression: client or server?

Anatoly Vorobey
Wed, 13 Aug 2003 22:59:05 +0300

On Wed, Aug 13, 2003 at 12:53:20PM -0700, Brad Fitzpatrick wrote:
> I agree.

Me too.

I'll mention in passing that one benefit of server-side compression you
haven't mentioned is that for some methods of text compression which use
dynamically built dictionaries the server is in a position to offer 
better compression rates, beause it sees much more data than the client.
I don't this this overweighs the other arguments in favor of client-side
compression though.

> So, yeah:  compression definitely client-side, but the server should do
> stats by the client sending more data about how much it shrank it to.  See
> any problem with that?

Only the fact that we're wasting 4 more bytes per item just for the sake
of compression stats. Some people may not be very interested in stats

Or do you see it as a system in which the 4 bytes are only stored for 
item in which clients sent compressed size, and not stored for other 
items? Possible, but will necessarily make the code more convoluted, and
add a slight CPU penalty, probably insignificant.