Re: modules

Terence Kelly wrote:
>
> I'm a Web caching researcher, not a Squid developer. I gather that
> the Squid dev team is discussing design issues for the next major
> release. Two suggestions:
>
> 1. Modularize in such a way that any reasonable replacement
> policy can easily be installed. Recent research suggests
> that policies such as GreedyDual-Size and variants of LFU
> offer substantial advantages over LRU-based algorithms.
> Squid will be a better product if admins can decide the
> replacement policy at install/configuration time; it shouldn't
> be hard-wired.

I'm an engineer, not a researcher, so my inclination is to do exactly
that, as I am unfit to evaluate high-order algorithms. I leave that to
people with the necessary mathematical backgrounds.

>
> 2. In the access logs, could you include a hash/checksum of the
> requested URL? My work depends on access logs, and it would
> help me to distinguish between cacheable/dynamic content if
> I had a hash of the data payload (excluding HTTP header).
> A last mod time for each URL would be useful too. As things
> now stand, when I see requests for "http://foo.com/" in the
> access logs with slightly different transfer sizes, I can't
> know for sure whether the document has changed, or whether
> the HTTP header has changed.
>
> Squid access logs such as those collected & published by NLANR are
> extremely valuable to researchers like myself. They'd be even more
> valuable if the one or two low-cost changes above were included.

This is a place where we _can_ definitely benefit from a module, I
think. Add an abitrary 'datum' to the access log, and allow something (a
module) to put whatever it likes in that datum. (of course, a module
could just write out information itself..but - from some rather
frustrating times hacking on squid - it all depends on if the
information you need is available at the place where you are working or
not) You would need to provide a number of points at which a module
could be attached to make it useful, of course, but I can see
considerable debugging/research/profiling benefits from having a tap at
various points in the sequence of events.

D
Received on Tue Jul 29 2003 - 13:15:58 MDT