squid cache replacement policy update

        I just submitted to squid-dev@ircache.net a patch (for the
squid-2.2.STABLE3 source base) that implements a set of enhanced cache
replacement policies. These policies use frequency of reference instead
of recency of reference to make replacement decisions. In simulation
and under synthetic workloads they achieve higher hit rates and/or byte
hit rates than the traditional LRU replacement policy. Not only that,
the execution of these policies in Squid consumes fewer CPU resources
than the previous LRU policy. We have several Squid caches using these
policies up and running in our lab and have had good results.

        I presented this as a work in progress at the recent Web Cache
Workshop in San Diego. A paper describing the work was included in the
proceedings given to participants. I have updated the paper to include
our latest performance characterization; it is available on the web at
http://www.hpl.hp.com/personal/John_Dilley/caching/wcw.html. A full
description of the replacement policies is also documented in an HPL TR,
http://www.hpl.hp.com/techreports/98/HPL-98-173.html.

        If you were at the workshop, or are just interested in how an
alternative replacement policy can help improve Squid's performance, you
might want to check it out. I am hopeful that this work will appear in
a future version of Squid; there also is no reason not to implement the
policies in earlier versions of Squid (which we did during development
and testing). I'm interested in feedback and comments, and can make a
copy of the modified source available to interested parties not on
squid-dev if there is sufficient interest.

        Best regards,

                             -- jad --

        John Dilley <jad@hpl.hp.com>
        Hewlett-Packard Laboratories
        1501 Page Mill Road MS 1U-17
        Palo Alto, CA 94304 // USA

        http://www.hpl.hp.com/personal/John_Dilley/
Received on Thu Jun 10 1999 - 22:42:32 MDT