[4921] in linux-scsi channel archive
Re: Quick overview of Disk Sharing (Was: Re: dual-attached disk)
daemon@ATHENA.MIT.EDU (Stephen C. Tweedie)
Fri Oct 16 17:08:43 1998
Date: Fri, 16 Oct 1998 12:08:27 +0100
From: "Stephen C. Tweedie" <sct@redhat.com>
To: physmsa@cantua.canterbury.ac.nz (Mr M S Aitchison)
Cc: linux-scsi@vger.rutgers.edu
In-Reply-To: <199810142300.MAA19813@cantua.canterbury.ac.nz>
Hi,
On Thu, 15 Oct 1998 12:00:48 +1300, physmsa@cantua.canterbury.ac.nz
(Mr M S Aitchison) said:
> "Failover" type of sharing has been around for many years
Yep.
> I think the hardware that makes it easy is a RAID controller with
> enough channels (or whatever the jargon) to connect to two computers on
> different SCSI cables.
Not necessarily. Most Unix solutions also directly support
multiple-initiator scsi (ie. a single scsi disk with two hosts on the
bus). Obviously, you need scsi driver support for this to work, but
it can be done. Currently this is the only supported configuration
for Microsoft's Wolfpack, for example.
> I also recall hearing of disk subsystems that not only can connect to
> multiple computers but to the ethernet directly - but I only have fuzzy
> ideas of what they claimed could be done simultaneously.
Under Linux, we can already export block devices over the network
using the nbd block device. :)
> However there a couple of totally different approachs, such as used by
> Sun with their cached filesystem and automounts.
The main advantage of this mechanism is a cluster-wide disk namespace
provided by NIS automount maps. Redundancy is there, but it's not
guaranteed.
> I think Linux could take both ideas, plus the idea of RAID or
> logging changes/journalling, and implement something more reliable
> but still aimed at systems doing many more reads than writes.
CODA already does this.
> But to do justice to this it probably needs a lot more than just a
> beef-up of an automounter but something like IBM?/Transarc
> Distributed File System
DFS and CODA share the same heritage. They both come from AFS, the
Andrew File System developed at CMU. Transarc bought AFS, but CMU
continued to develop CODA. CODA support is built in to 2.1 kernels.
CODA supports lots of goodly things. It includes local caching of
files, proper distributed locking of updates, server replication and
automatic reconciliation after data loss, and all sorts of stuff like
that. However, it is not suitable for situations where you have files
open for shared write. You still need something like NFS for that,
so that all of the locking and write consistency is properly managed
by a single centralised server. Of course, there's nothing to stop
you from NFS-exporting a CODA-replicated filesystem if you wanted, and
then you can failover the NFS server whenever you need to!
Nice, isn't it? :)
--Stephen
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
