Re: Calculation of age headers (AGE-CALCULATION last call)
Jim Gettys (jg@pa.dec.com)
Tue, 11 Nov 1997 08:29:08 -0800
Consider this a LAST CALL on:
http://www.w3.org/Protocols/HTTP/Issues/#AGE-CALCULATION
This hot potato has ended up in my plate despite repeated attempts
to avoid it :-(.
I read through both Roy's and Jeff's Internet drafts on this topic,
and all the mailing list mail on the topic yesterday, to come to some
decision on the topic, since it hasn't been obviously settled despite
many attempts over a long time.
There is a (very) rough concensus supporting Roy's position in the mailing
list, in particular that Jeff's option (c) would cause non-caching proxies
to have to do onerous implementation work; in fact, one could argue that
adding another clock to the mess would just make things worse. Such proxies
are often used as part of a firewall complex (one tranparent proxy at the
firewall itself, with another caching proxy just inside; this avoids having
a big, complicated, caching proxy on the front lines of a firewall; the
more code in them, the less they are to be trusted).
While I think that Jeff is technically correct, I think a pragmatic attitude
needs to be taken on the implementation cost/benefit side; if you have an
HTTP/1.0 proxy up stream of you, you are in very serious trouble in the
first place, so I've decided to add to the very rough concensus already
on the list (making it rough concensus, I guess; not all that many people
have actually commented).
I'm currently leaning against Koen's suggestion clarifying that
such proxies MAY add age headers, as much on complication grounds
as any other (things are complicated enough as it is). Koen's mail message is
http://www.ics.uci.edu/pub/ietf/http/hypermail/1997q4/0002.html
There were no replies to Koen's mail on this topic, so I have no other
opinions to guide me. Other opinions on this suggestion are solicited
(particularly from implementers...).
If others have opinions on either the AGE-CALCULATION issue, or on Koen's
suggestion, NOW is the time to speak up. My current plan is to adopt the
language in Roy's draft
(ftp://ds.internic.net/internet-drafts/draft-fielding-http-age-00.txt). I
consider the wording to encourage synchronized clocks an editorial issue,
in an implementation note.
- Jim Gettys