Re: Mail to squid-dev

[forwarded to me because I had dropped off squid-dev sometime ago..]

> > One thing that has been discussed is the possibility of moving the store
> > heap code out of the store routines and making it per-fs . That way,
> > each fs has its own object replacement algorithms, which squid can 'alter'
> > (though storeExpire() calls? Its been a week or so since I've looked at this
> > particular bit of code, my hands are dirty elsewhere right now). You might
> > be asking "Why?", but consider how you'd integrate a rotational filesystem
> > in the current framework ..
>
> In fact, this is exactly what I did. COSS (btw, COSS = Cyclical Object
> Storage System) is indeed a rotational filesystem. As such, there is no
> need to expire objects from it at all.. Objects are stored in LRU order on
> disk, as new space is allocated old objects are simply overwritten.

Interesting. So does your squid actually support both filesystem types?
If so, I'm curious to know how you abstracted the heap code from an fs
layer point of view, because after I get some more work done on my
current hack run, I'm going to work with Arjan and Henrik on doing
something similar, and releasing another patch.

Adrian

-- 
Adrian Chadd					Systems Engineer
<adrian@ip.versatel.net>			Versatel Telecom BV
						Amsterdam, The Netherlands
	"Music in the soul can be heard by the universe" - Lao Tsu