Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 17 Sep 2003 14:33:11 -0500
From:      Dan Nelson <dnelson@allantgroup.com>
To:        Eric <esmith@jaguar.ir.miami.edu>
Cc:        freebsd-questions@freebsd.org
Subject:   Re: fbsd & fibre channel & SANs
Message-ID:  <20030917193310.GF48979@dan.emsphone.com>
In-Reply-To: <Pine.OSF.4.56.0309171422270.17091@jaguar.ir.miami.edu>
References:  <Pine.OSF.4.56.0309171422270.17091@jaguar.ir.miami.edu>

next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (Sep 17), Eric said:
> I'm trying set up a cluster of freebsd fileservers sharing a common
> set of volumes/drives/raids over netatalk to a bunch of OS 9 clients.
> Each fileserver has a Qlogic 2100 fibre channel card connected to a
> switch. Also connected to the switch are several fibre channel raid
> controllers, with raids attached to them. I have no problem mount and
> using the raid volumes on the freebsd servers.
> 
> What I'd really like to do is mount the same volume on two or more
> servers, which I can do, but one server does not know about the other
> server's changes to the disk. I'm sure this results in files being
> overwritten since each server thinks it owns the disk. I'm sure
> things like softupdates compund my problem even more.

You would need a shared storage filesystem; GFS is the only one I know
of, and that's Linux-only.

> Is there a way to have two machines share a disk, and communicate
> between them (over the network?) the changes? I've seen AFS, and it's
> mostly what i'm trying to do, but it doesn't work with OS 9 clients.
> I want an OS 9 client to be able to pick a server in the cluster
> (from their chooser) and be presented with the same shared volumes,
> regardless of the chosen server. Any help??

AFS looks like it replicates files onto multiple servers, so if one
goes down the data is still available somewhere else.  The servers do
not share backend filesystems.

The best solution today would be to have one master server that mounts
all the volumes (with softupdates, but short cache times), and one
standby server that polls the master and as soon as it stops responding
to pings, mounts the filesystems, ifconfigs an alias IP to take over
the master's IP, and kicks off background fscks to clean up the
filesystems.

If the servers are stable, you can probably spread the filesystems
across multiple servers, then cross-nfs-mount everything so the
netatalk clients see the same set of volumes on each server.  Some will
end up going over NFS to get to the server really mounting the
filesystem, though, so make sure your server-server links are
high-bendwidth.

-- 
	Dan Nelson
	dnelson@allantgroup.com



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030917193310.GF48979>