Re: a patch for coss

I've no objection - but would like to see the patch :}.

Rob

On Fri, 2003-05-23 at 05:04, Duane Wessels wrote:
> I spent some time re-familiarizing myself with COSS and running
> some polygraph tests. Since COSS uses 'e->swap_filen' as the offset
> to the storage file, and because sfileno is signed, we are limited
> to about 2^24 bytes, or 16M.
>
> In a polygraph test with cache_dir larger than this, Squid asserts with
> an error because e->swap_filen becomes negative.
>
> If, instead, swap_filen becomes a the index to, for example, a 512
> byte block, we can get up to about 8GB. Larger block size obviously
> leads to larger total file size. The downside is, of course, that
> we'll waste some space at the end of each object.
>
> I have this coded up and it seems to be holding up under polygraph,
> although the hit ratio degrades over time and I see some "failed
> to unpack metadata" messages.
>
> I would like to commit this, with the COSS author's and user's approval...
>
> Duane W.

-- 
GPG key available at: <http://users.bigpond.net.au/robertc/keys.txt>.