Re: "Hits" pragma

Brian Behlendorf (
Tue, 15 Aug 1995 16:13:55 -0700 (PDT)

On Tue, 15 Aug 1995, Roy Fielding wrote:
> Paul said:
> >I don't think externally routed reports create the proper
> >incentives for wide adoption -- it's a fragile system that requires
> >too much advance cooperation. In contrast, the forwarding of
> >"bundled requests" upon expiration requires no additional
> >cooperation between servers and proxies.
> That's true, but it leaves out the following:
> 1) Most servers won't care about that data, so it is sending
> unnecessary information.

I disagree - I think *all* servers would be interested in that data. Who
do you know who runs a server and doesn't care how many hits they get?

> 2) The last action of a cache does not involve a request to
> the origin server, so there is no request on which you
> can bundle the last set of information.

This is true. Perhaps at cache-flush time some sort of bulk reporting
mechanism can take place, where every URI at is reported
on in one transaction to Paul, want to revise the
proposal? Yes, it unfortunately is one of those "batch reporting"
mechanisms we'd like to avoid, but if the cache flushing is done often
enough it might not be hard, and the justification for asking caches to
perform this would be "to be a responsibly cooperative net.entity".

> Keep in mind that the proxy may know a lot more about an individual
> client than is available via HTTP. For example, the proxy may have
> authenticated the user and can generate its own anonymous id. So, I
> would say:
> timestamp HT domain HT [anonymous-id] HT [referer]
> would be just fine, with domain being defined by the proxy.

What's HT? and is the record above on a per-hit basis or aggregate? If
the former, I would be happy with this. anon-id and domain don't have to
map to what they "really" are. Have we found a common ground?

> Can this discussion be moved out of HTTP-WG now? I would not
> want to attempt standardization until it can be tested on at least
> one implemention, and this discussion is drowning out the real work
> that we need to accomplish RIGHT NOW on the 1.0 and 1.1 specs.

If Apache had a proxy-server module, I'd dive in and do it :)

Ari? What do you think?


--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-- http://www.[hyperreal,organic].com/