> Date: Tue, 8 Jun 1999 11:39:27 -0400 (EDT)
> From: Jeff Garzik <jgarzik@pobox.com>
> To: Christoph Lameter <christoph@lameter.com>
> Cc: Juan Antonio Martinez <jantonio@dit.upm.es>,
linux-kernel@vger.rutgers.edu
> Subject: Re: cachefs module for linux?
> On Tue, 8 Jun 1999, Christoph Lameter wrote:
> > On Tue, 8 Jun 1999, Jeff Garzik wrote:
> > > On Tue, 8 Jun 1999, Juan Antonio Martinez wrote:
> > > > 1- Are there any fs module to implement a local disk cache over
> > > > NFS readonly mounted filesystems? Something about Solaris's cachefs
> > > cachefs gets mentioned now and again, and is very useful. We used it on
> > > the cnn.com Web farm, where we had one big machine (NFS server) and
> > > several Web servers all mounting the big server's drives with
> > > nfs+cachefs. cachefs really improved throughput, but cnn.com uses
> > > almost exclusively static files.
> > squid does the same in a far more elegant way as we implemented for
> > the Sony Motion Pictures site. NFS is the wrong solution and the wrong
> > protocol for web caching.
So you suggest something like webNFS?. Perhaps is an action to study...
> True, but Squid didn't exist, or wasn't well known in 1996.
> > I thought NFS already did some caching?
> Not enough for the CNN Web site, which is probably at least half a
> terabyte of stories by now. The large selection set of
> frequently-viewed pages demands something more than simple buffer
> cache.
> cachefs comes in really handy, too, setting up "diskless" machines with
> nothing but NFS cache and swap space.
>
This is my problem: I manage a students lab with 100+ NFS clients
and 4 binary apps NFS servers. Every client has a small ( about 400MB )
hard disk to store a tftp retrieved mini root filesystem ( about 20 MB )
and a swap. /usr /home, and so are nfs mounted. Specifically /usr is readonly
and need to manage large apps ( Xserver, java compiler, netscape, and so) in
a more efficient way than NFS cache offers ( some clients has only 16Mb RAM )
You'll say: use coda . OK but our feeling is that coda is still
too unstable to manage about 1500+ users and 100+ clients in an "hostil"
environment such a students lab. Item more: we need quotas, file locking and
so that are still not supported in coda ( needed for the /home directory )
So a hard disk cache for readonly netmounted filesystems seems the
ideal solution for us. Of course: using knfsd, 100Mb network, load balanced
servers, etc... neither servers nor network are now the problem. Just need
to improve eficiency in clients
Juan Antonio \|||/
/ _ _ \
\ o o /
=========================o00o===U===o00o======================================
Juan Antonio Martínez Universidad Politécnica de Madrid
email: jantonio@dit.upm.es E.T.S.I Telecomunicación
http://www.dit.upm.es/~jantonio Ciudad Universitaria s/n
Tel: 34-1-5495700 ext 328 Centro de Cálculo
Fax: 34-1-3367333 Madrid, Spain
==============================================================================
Allways look for the right side of life - Monty Phyton en "La vida de Brian"
-
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/