Re: Caching dynamic pages (CGI) for one service/server

Nottingham, Mark (Australia) wrote:

> I had a nice conversation with Danzig at the WCW about this, and he said
> that he's moved to using presence of a validator (practically, LM) as an
> indication of cacheability (they've left their equivalent of no_cache in
> just to give warm fuzzies to some customers, although IMHO it's worse than
> useless). They did some testing, and found that they lost only a small % of
> cache objects, but coherence went way up. It makes a lot more sense than
> pattern matching the URL, whether through no_cache or refresh_pattern.

I think the general consensus of the Squid developers is the same.

The main purpose of refresh_pattern is to provide different staleness
levels for different kinds of cachable objects, not to enable caching of
dynamic pages. The ability to specify a min age is ONLY for explicit
caching "dynamic" objects and should/must not be used on general
patterns.

no_cache is sort of a short hand for denying caching of such objects
that one do not want in the cache, like local sites and other things
where caching is more or less worthless or freshness is critical and one
happily pays for the bandwidth/latency required to fetch the most up to
date contents (for example the local newspaper or stock market, where
the local freshness policy is stricter than the content providers)

--
Henrik Nordstrom
Spare time Squid hacker