My point about counters was that they are indicative of the desires of
information providers -- to know how many accesses they've had and to make
use (however frivolous) of the count. I think counters are a good example
of this impulse because they immediately share the data with visitors
(i.e., make use of it), but other server administrators may be just as
satisfied with less immediate or public hit counts. I meant counters as a
good 'counter' to Roy's statement: "Hits received on the proxy are amusing,
but not necessary for anyone operating a non-marketing-oriented service."
Should the word 'necessary' be weighed against the necessity of a home page
in the first place? Many amateur information providers --
non-marketing-oriented services, just fans and the like -- place great
value on hit counts.
>Finally, for those who want the counters on those pages probably would
>not be satisfied at all with any kind of statisically-based proxy
>cache reporting. In particular, any scheme where counts might get lost
>or even delayed a few days would interfere with the goals of those who
>would be counted. Pages with counters *can't* be cached, because the
>content changes every time it is accessed. So your anecdote is
>contrary evidence, not supportive, of the proposals that the proxy
>servers should accumulate statistics and send them back at some later
>time.
I agree with your point if not your conlcusion. Again, I meant to use
counters as a particularly visible example of the desire for counts, not as
the only case of it.
>Do you have any evidence (anectodal or more formal) that there are
>web resources for which:
>
>(a) it is possible to cache the resource and
>
>(b) unreliable delayed measurements of accesses (until the proxy next
> reports them) are acceptable
Sure, all static pages fit (a) & (b).....
>(c) statistical measurements (measuring accesses every 8th day)
> are not acceptable
>
>If there such resources, are there sufficiently many to warrant the
>extra overhead of statistics gathering in the cache service dominates
>the extra overhead of just retransmitting the data each time?
Well, I don't have a problem with statistical measurements (c), so I won't
argue against their use, in substitute for a 'hits' report, as a distinct
isssue. I understood, however, that one aim of this proposal was to
encourage proper use of 'expires', and to entice such proper use through
accurate, though delayed, reporting. This seems a practical proposal to me
-- one that saves the provider server resources, and allows the proxy less
volatility in its cache.
If your position, though, is that the spec is the spec, 'Expires' should be
used as written, samples are plenty good enough for hit counts, and
enticements be damned....well, okay. I just don't think that's as
practical a proposal if the goal is to change providers' behavior.
Marc Hedlund <hedlund@best.com>