Re: Disk I/O

On Sat, 2002-10-12 at 02:12, Henrik Nordstrom wrote:
> On Friday 11 October 2002 16.09, Денис Овсиенко wrote:
>
> > I absolutely agree on index things (see above), but it naturally
> > comes to storing objects' body in the same database to not
> > duplicate file sizes and to use the same way to extract them.
> >
> > P.S. It's just an idea, that seems useful to me and I wished to
> > donate it here.
>
> You are more than welcome to help and try out the idea and prove Joe
> and others wrong here.
>
> However, one immediate problem is the current code requirement of a
> in-core cache index (there must be no delays in index lookups), but
> the refactoring of storeGetPublicByRequest() Robert is doing aims at
> making of it possible to have a "on-disk" index stored elsewhere than
> in the memory of the Squid process. The Squid development team has
> the long-term goal that a "object store" should be able to choose
> freely how it implements both storage and indexing.

Further to this, we have a sourceforge project dedicated to
experimenting with squid - devel.squid-cache.org is the homepage. If you
have, or create a sourceforge developer account, we will give you write
access to create your own branch and develop your idea there. Doing it
on the devel CVS tree allows you to merge HEAD changes in quite easily,
and to generate patches for submission to HEAD easily too.

Cheers,
Rob