Re: Webbrowser cache implementation.

> I was considering implimenting a sort of pre-caching algrithim where the
> browser could pre-load pages off of links. Obviously this raises serious

This does not have to be and probably should not be done in the browser but
in the cache itself. In fact putting a great degree of caching in a browser
at all doesn't really make much sense. What would be more useful (assuming
you're talking about telephone line slip/ppp users here, ie not a fast
connection) would be to write a good general proxy like squid that runs
on the client PC, independent of the browser. Obviously squid itself
is too memory hungry, but a client for a personal machine can trade
off a lot of memory for CPU time, because all it has to beat is the
speed of a serial line.

Doing prefetching with squid is relatively easy: you find out which pages
are fetched by users from monitoring the log output in real time *on the
cache server itself*. You request the same pages the users requested,
parse them and then fetch the next level of pages. These prefetches are
shown in the logs as having been requested from the cache server, not from
one of the clients, so the process looking at the log files can ignore
its own requests and not get stuck in an infinite recursion without
needing any fancy recursion detection logic. Only leaf caches
should do prefetching, so that the prefetching stops at one level.

Also, you should only prefetch from slow sites. Why bother with fast
ones when you can get it direct anyway. Of course, this may make slow
sites slower, so maybe there's no point in the exercise at all :-)

Graham
Received on Wed May 28 1997 - 20:04:53 MDT