On Wed, Jul 25, 2001 at 05:32:39PM +0200, Przemyslaw Wegrzyn wrote:
>
>
> On Wed, 25 Jul 2001, Henning Brauer wrote:
>
> > On Wed, Jul 25, 2001 at 05:16:44PM +0200, Przemyslaw Wegrzyn wrote:
> > > On Wed, 25 Jul 2001, Henning Brauer wrote:
> > > > Clustering is a bit misleading for this qmail-ldap technique. You want a big
> > > > NetApps Filer and mount the users maildirs on each machine via NFS.
> > > NetApp Filer is too expensive for us right now, but we'll probably buy
> > > one if the system grows :)
> >
> > Big systems cost big money.
> > The only alternative is a BSD based filer. Linux NFS is, eh, .. wel, you get
> > the idea.
>
> BTW, does *BSD support Mylex AcceleRAID 170 or Adaptec 2100S ?
> I have not much experience with BSD yet.
> One more off-topic question - what are the problems with Linux NFS
> implemetation ? I've heard about locking issues...
>
> -=Czaj-nick=-
*BSD supports Mylex and AMI Megaraid products. Stay away from Adaptec
crap. It's actually DPT, and DPT is AFAIK slowwwwwww.
Also, ICP-Vortex makes good RAID controllers.
I just bought a Mylex eXtremeRAID 1100 with 16 MB
of cache and two channels for $250 on Ebay! The card retails for over
$1000. The proprietary cache and battery backup module alone cost
like a RAID controller itself in retail.
You will definitely get a better deal if you buy a Megaraid
controller (upgrade-wise and most likely performance since Megaraid
controllers do not use proprietary memory). My choice for Mylex is
simply because Linux and NetBSD and FreeBSD have all support for it.
Many Megaraid controllers are supported by Linux and FreeBSD, but not
NetBSD.
I would stay away from NFS at all costs. It simply sucks. I don't care
how expensive your filer is, it still sucks, sucks, and will always suck
in my mind for any purpose. Putting a filesystem over a network layer
for clustering is a design flaw. You never know what bug you'll find next
in any NFS implementation. Each server in a cluster should be connected
directly to a redundant shared array. If you can't afford a shared, keep
a separate small array per server. Sure your management will be more
difficult with separate arrays for each server, but at least you'll save
yourself any possible NFS headaches. GFS for Linux and shared fiber array
is THE solution for a mail cluster, not an NFS box.
