- From: Balint Nagy Endre <bne@bne.ind.eunet.hu>
- Date: Sat, 12 Aug 1995 12:06:09 +0200 (MET DST)
- To: burchard@horizon.cs.princeton.edu
- Cc: http wg discussion <http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com>
Paul Burchard writes: > Brian Behlendorf <brian@organic.com> writes: > > One thing that would encourage the use of Expires: > > headers of course would be a way for caches to report hits > > they served directly without a long-distance conditional GET. > > How about using an additional pragma directive > > Pragma: hits = 35 > > to denote the accumulated number of unforwarded requests received > by a proxy, in addition to the request being forwarded. Multiple > "hits" directives make sense, and can be accumulated by intermediate > proxies. I'd preferred a Pragma: hits hits# proxy-name variant instead. > Using this pragma, hit counts will be accurate over time intervals > on the order of the expiration interval of the resource (which is > tunable by the server). > > This proposal requires only minimal changes to caches, servers, and > log analysis tools -- and offers a graceful, incremental upgrade > path in the meantime (since Pragma headers are already passed > through by any conforming proxy). Performance of all components is > practically unchanged since no additional network connections are > used. And perhaps most important for successful adoption, this > scheme avoids imposing any burdensome reporting duties (such as > "accounting batch runs") on proxy maintainers. Agree. Before I read this, prepared a discussion of hit reports, now I escaped finishing it and you escaped reading it. Without the Hits pragma, some web-admins or document owners would use the no-cache pragma to have those hit counts, which has a very expensive side effect for both the client and server sides. (bigger system load on server, and bigger costly wan traffic for both.) But will the web-admins and document owners be satisfied with raw counts? And on the opposite end, will admins running caching proxies on firewalls honor specs for more detailed statistics? We shall found a good compromise between the stats hungry server-side and terse client-side requirements, otherwise too many misconfigured systems would be running in the near future. (I mean possible mis-use of no-chache pragma, and consequently caches not honoring it.) Now that we have this idea intoducing the hits pragma we have a lot to do elaborating the concise definition and exact usage of it, and we shall have to finish it till 18 Aug. I presume we shall incorporate the hits pragma in the ongoing version of the HTTP draft as a must requirement. Any ideas, suggestions, comments? Andrew. (Endre Balint Nagy) <bne@bne.ind.eunet.hu>
Received on Saturday, 12 August 1995 03:24:46 UTC