Re: idea for making better the binary tree code for ACL's

>
> I have an idea for making better the ACL search code.
>
> In situations where you have thousands of ACL's, putting
> them into a binary tree (or a splay tree) should speed up the ACL
> matching. It was brought up, however, that binary trees are inefficient
> for the typical dataset that comprises ACL's.
>
> What if you were to compress the url?
>
> What if you were to use a dictionary based off of thousands
> of random url's, store this dictionary staticly in the code, and
> use this dictionary to compress the url before storing it?
>
> Would removing some of the redundancy in the url make it
> faster to find?
>
It's a similar idea to one that's been brought up before (Mike OReilly from
iinet), perhaps it'll appear in 1.2.

It would nice to compact all those "http://www." prefixs into a dictionary
lookup. Hopefully a substantial memory saving will result as we find the
maximum proxy server is NOVM + 128M RAM + 16Gig of cache. Anything larger
and Squid drops into swap. Unfortunately 64Meg SIMMs aren't very cheap
yet in Australia - cheaper to get a PC motherboard with more SIMM slots
for 32s.

tom@interact.net.au
Received on Tue Sep 16 1997 - 23:40:45 MDT