devfs, was RE: NFS, again...
Thu, 04 Oct 2001 10:12:11 -0700
Sorry, prematurely sent that... I am lucky to have what I think will be a
If I am working with Volume 0 on my array, which I want mapped to several
LUN 1 0 0 Effectively /dev/sdb on node1 and
/dev/sda on node2
LUN 2 Empty 0 Effectively /dev/sdb on node2
LUN 3 Empty Empty
... ... ...
I suppose that most any DAS RAID array will do this. I think this will
work, which proves the potential value of using a shared DAS RAID array in
an HA cluster. Hopefully the major/minor numbers will match up, and there
will not be any more bumps on this road.
From: Upton, Sean
Sent: Thursday, October 04, 2001 10:06 AM
To: 'James Bottomley'; firstname.lastname@example.org
Subject: RE: devfs, was RE: NFS, again...
Fortunately, in theory, my RAID array's volume-to-LUN mapping tool is farily
flexible, allowing my to map volumes to LUNS on each host. My mapping table
would look like:
From: James Bottomley [mailto:James.Bottomley@SteelEye.com]
Sent: Thursday, October 04, 2001 9:54 AM
Cc: James.Bottomley@steeleye.com; email@example.com
Subject: Re: devfs, was RE: NFS, again...
> Actually, my second node has /dev/hda1 as its root device (Ultra 10s
> are cheap IDE boxes) and only has one SCSI device, which is this
> single volume from my drive array. So using /proc/scsi/scsi might
> work if I had another direct-access device or volume to pad /dev/sda
> with so I can add my array volume as /dev/sdb1? Or am I
> misunderstanding? I might be able to do this by recompiling my kernel
> with multiple-LUN support (which is currently off for simplicity's
> sake), and changing my LUN mapping parameters on my DAS/RAID array,
> which is configurable in this regard, but I am wondering if there is a
> more simple and/or graceful solution.
I think I reversed your nodes. The node which has a real /dev/sdb1 as /data
is node1 not node2? In that case, If you can reverse /dev/sda and /dev/sdb
node1 then use /dev/sda1 on node 2 everything will work.
Alternatively, as you suggest, if you get an extra LUN to show up on node2,
you would then be able to swap this with /dev/sda, so the entire cluster
now use /dev/sdb1 for the real device of the NFS volume.
The trick is just to make the real device the same on both nodes.