Re: Alternative Squid Filesystem

Found this lying around in a pool of old emails, not sure if I have
responded.

Your ideas seems very similar to the ideas I was toying around with some
year ago. Part of it has been implemented by Eric Stern in the COSS
object storage available in the current Squid CVS sources.

For an archive of old notes see http://squid.sourceforge.net/hno/

--
Henrik Nordstrom
Squid hacker
Alexander Pennace wrote:
> 
> Greetings,
> 
> Over the weekend I've been toying around with a different approach to
> managing Squid cache storage.
> 
> The main problem appears to be keeping fragmentation down while still
> maintaining a reasonable LRU replacement policy. If we say that the
> replacement policy doesn't have to be a perfect LRU policy, then a
> number of interesting options open up to capitalize on that sacrifice.
> 
> Imagine a Squid spool as a large linear file. This file is made up of
> contiguous objects for the cache. Adding new objects to this spool is
> simple: if there is enough free space for the object add it there.
> Otherwise consider each object or contiguous group of objects that if
> freed will yield enough space for this new object, and replace the
> object or group of objects with the highest average used-age.
> 
> Thats essentially my idea. I'm still considering other additions, like
> maintaining timing data to get an idea of which contiguous blocks from
> disk are the fastest to read or write to, but those don't directly
> impact my basic theory.
> 
>     ---------------------------------------------------------------
> 
>                Part 1.2   Type: application/pgp-signature