September 9 at 11:08 am
- via Bookmarklet
- Link
Elias Torres, Amit Patel, Derek Collison and 17 other people liked this
+ one un-like for groups requiring me to sign in - Sanjeev Singh
Very glad to see this published. - DeWitt Clinton
The link to the paper: http://sdch.googlegroups.com/w... - DeWitt Clinton
the ppt says 40% compared to gzip, which is really great. - Christian Sonntag
I like this. I'm kind of confused by the domain/path part though. I was about to complain that they were copying the problematic Set-Cookie header, but then none of the examples actually include any kind of path or domain info. It would also be nice to require the content hash in the dictionary url to eliminate any update or caching questions there (new dict requires new url). - Paul Buchheit
I would also like a variant where the client simply says "always sdhc" and the server simply picks the best dict and includes the url of the dict in a response header. This would occasionally increase latency a bit (if you didn't have the dict), but it would be easier for the client to implement because you would cache the dict the same as any other resource and not have to think about keeping a list of candidate dicts. - Paul Buchheit
Also, why not allow differential vs a previous etag? It seems like a straightforward extension, and of course servers would be free to ignore it. - Paul Buchheit
Paul, as written the client sends a hash of each dict, so if the server send a get-dictionary, that means the client's dictionary is stale - Casey Muller
Yes, when everything works, but if you have several servers and they get out of sync, or maybe there's a dumb cache in between, it will cause problems. In general, life is much simpler if new content gets a new url. - Paul Buchheit
Yeah, definitely. It also doesn't help that Example 4 in the doc is wrong, since it still has a get-dictionary in the server's response - Casey Muller
How many servers actually use gzip? - Brian Sullivan
Yes, gzip is inexcusably unused by many sites, but many of us do use it and it makes our sites faster :) - Paul Buchheit
Paul -- any statistics on use that you know of? - Brian Sullivan
I guess what I'm really looking for is to have the protocol simply say "this page is a delta of X", and X could be a previous version of the page, or it could be the URL of a site-specific dictionary. That would make it very nice for feeds or feed oriented pages obviously :) (and that includes a large fraction of the non-search content, such as news) - Paul Buchheit
HTTP rsync encoding? http://rproxy.samba.org/doc/pr... - Mark Trapp
yeah, "delta" or "patch" on the client side would make a real impact. - MikeAmundsen
It depends how you measure it Brian. I don't know what fraction of sites, but I think most major sites support it (Google, CNN, Facebook, and FriendFeed, for example), so a large fraction of pageviews are compressed I expect. - Paul Buchheit
On ground this looks so simple, that I am amazed that nobody thought of that before. - Varun Mahajan










