RE: Fw: Proxying PHP Pages

[ I haven't replied to isp-users, as I'm not subscribed- feel free to
fwd ]

> The Squij work looks interesting. How do you tell which hits
> were for stale objects? Does it use a single cache log, or logs from
> multiple servers, unified to detect modifications? In our experience

Single cache log. With squid, you can tell if a hit was fresh or stale
by a tag in the log; see the source for a definative list.

> it's been very hard to identify modifications by looking at
> logs (Squid
> or otherwise). You can not detect multiple modifications
> between hits,
> so that causes you to underestimate modifications. On the other hand,
> if a document is kicked out of cache the log will show a 200 and it's
> hard to determine if it was a modification or a capacity miss ...

In this context, Squij gives a metric of how many stale hits were
modified vs. unmodified on the origin server. By it's very nature, Squij
is imprecise; its only view of the data is through the logfiles.
Therefore, it cannot give definative freshness or modification
statistics. However, it can give this information *relative* to past
performance, broken down by refresh_pattern; this should be enough to
optimise the pattern.

The difficult part, of course, is validating this assertion; I'd welcome
any thoughts.

> A reference for Alex is Cate's paper, "Alex - A Global File
> System" in Proceedings of USENIX File System Workshop May
> 1992 (pp 1-12)
> and briefly in Gwertzman and Seltzer's USENIX paper "World-Wide Web
> Cache Consistency" http://www.eecs.harvard.edu/~vino/web/usenix.196/.
> Also Krishnamurthy and Wills USITS paper on Piggyback Cache
> Invalidation
> at
> http://www.research.att.com/~bala/papers/pcv-usits97.ps.gz. >
Regards,

Thanks, will have a look. There were a number of relevant papers
presented at USITS, in particular, ones that give reference counts and
modification profiles for different kinds of objects; this is what I'm
basing my work on.
Received on Tue Dec 15 1998 - 21:05:21 MST