Re: "Hits" pragma (objection and alternative)
Balint Nagy Endre ([email protected])
Fri, 18 Aug 1995 03:56:36 +0200 (MET DST)
Reinier Post <[email protected]> writes (on www-talk list):
> Paul Burchard write:
> >
> >Talking to several people privately, I've been convinced that
> >"bundled request" reporting information would most naturally be
> >placed into a Forwarded header, instead of a Pragma.
Again: this is eligible for sending bare hit counts, but
current servers won't go crazy seeing this?
> Excuse me for jumping in like this, but ...
>
> Instead of using a header, ordinary HTTP transactions can be used to
> query proxy servers for state information. Particular types of request
> will yield particular excerpts from the logfiles. The standardisation
> required would be in the URLs used to access this information. URLs like
>
> http://proxy.server/getlogs?format=ncsa1.4+type=access+from=01/01/95
> http://proxy.server/getlogs?format=ncsa1.4+type=proxy+from=01/01/95
> http://proxy.server/getlogs?format=ncsa1.4+type=cache+from=01/01/95
>
> might return the specified section of the logs (as far as they are available)
> in the specified format (if the format is supported), grepped for the
> host making the request. (This requires special attention, if the requests
> themselves are forwarded through proxies.)
>
> Something like
>
> http://proxy.server/getlogs?report_implementation
>
> might be employed to make the server return a free-format page containing
> list of supported formats, the range of data for which logs are available,
> an email address for further information, whatever.
>
> Servers, and clients as well, have the right to view the exact imprints they
> make in proxy logs. Why invent a special format allowing proxies to censor
> this information? Just serve the logs!
>
> Advantages:
>
> - user (server-side) initiated: 'piggybacking' this information onto
> document requests has several obvious disadvantages;
> - not restricted to proxies: this mechanism allows all servers, clients,
> and proxies to query each other for information;
> - open format specification: by using logfile formats,
> any selection of information can be used, provided that a logfile
> format is invented to support it; why invent a new, fixed, and overtly
> specific standard, if logfiles already contain the information?
> - doesn't require a change to HTTP, which is supposed to be a simple and
> standard protocol;
> - not restricted to HTTP servers and clients. use it on your ftp site :)
And one more advantage:
Caches running full http servers can be extended with CGI scripts to do this,
but there are specially designed caching proxies, too.
At least we shall intorduce some (request) header to advertise the server port
number, having the stats for such caches! (I mean a full server, running on the
same host as the proxy.)
But I have one serious con too:
Log files, made by current servers may contain information, which sholudn't
be sent to servers (see privacy related notes on http-wg list).
By theory, even web-admins shall gather long-term statistics by filtering out
some fields, and not send "raw" logfiles to net at all, I think.
> A Forwarded: header is necessary, if only to prevent looping, but
> I don't think it's the right vehicle for this type of information.
>
> Reinier Post [email protected]
> --
> a.k.a. <A HREF="http://www.win.tue.nl/win/cs/is/reinpost/">me</A>
Andrew. (Endre Balint Nagy) <[email protected]>