Re: OSDI paper - IO-Lite: A Unified I/O Buffering and Caching System
Jim Gettys (jg@pa.dec.com)
Wed, 3 Mar 1999 07:06:16 -0800
> From: Oliver Xymoron <oxymoron@waste.org>
> Date: Tue, 2 Mar 1999 23:41:28 -0600 (CST)
> To: Jim Gettys <jg@pa.dec.com>
> Cc: linux-kernel <linux-kernel@vger.rutgers.edu>
> Subject: Re: OSDI paper - IO-Lite: A Unified I/O Buffering and Caching System
> -----
> On Tue, 2 Mar 1999, Jim Gettys wrote:
>
> > It shows a new I/O approach that is very general and flexible, and avoids
> > data copies with minimal overhead, even between processes. The authors
> > use as an example Web service, and show very good performance gains.
>
> Very interesting. We probably wouldn't see quite the gains the author saw
> with his FreeBSD implementation. It's a good match for FreeBSD as it
> already has a zero-copy scheme in a lot of areas and fits better with
> their design philosophy. Linux largely avoids VM manipulations like those
> suggested for instance by I/O-Lite's IPC scheme, because of their nasty
> effect on CPU caches, SMP scaling, etc. So parts of it are going to be a
> hard sell. And we've got sendfile(2) now, which is not nearly as pretty or
> generic, but might still give similar performance at least for the common
> web server case.
>
>
Sendfile does not help the CGI case, which is very common (or dominant)
on many (most) high volume servers.
I believe it will also help the CPU cache case for CGI; it avoids having
to touch the data again in many operations, which would pollute the cache
with data that was merely copied in the web server.
- Jim
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/