Re: MemPools rewrite

"Chemolli Francesco (USI)" wrote:
>
> > And there currently is at least one implementation using "the right
> > approach", based on a variant of the reiserfs filesystem on Linux. In
> > there the whole cache maintenance is moved to the file system. Squid
> > only stuffs content there and retreives objects as needed.
>
> I sincerely hope that this will soon see the light.
> Hans Reiser promised that the source would be released,
> the question now is: when? I am quite interested in this
> development thread.

There is nothing secret about what's going on with reiser_raw at this
point. It just hasn't made it's way to a public CVS yet (though it
will).

I'll post some tarballs publically, along with instructions (meant to do
that a couple of weeks ago, but got too busy to think straight) for
compiling and using. Anyway, I'll take the heat for not getting the
code into a publically accessible forum in a timely manner. Hans is
just providing the programmers, he doesn't pay too much attention to
what goes on with regard to making the code available, so don't shoot
Hans. Shoot me instead.

The storetree code is really very effective, to change the subject to
performance. The Squid we used at the Cacheoff-3 had no swap.state in
memory at all. The Squid memory footprint was a measly 67MB (as opposed
to the 230MB we would have seen from 2.2STABLE5 with async i/o)! And
its pretty darn fast as is. Adding a cache-digest style in memory
hit/miss checker would probably add a good margin of performance,
without sacrificing all of the memory gains.
                                  --
                     Joe Cooper <joe@swelltech.com>
                 Affordable Web Caching Proxy Appliances
                        http://www.swelltech.com
Received on Thu Nov 02 2000 - 01:47:46 MST