[linux-lvm] lvm in linux SAN enviroment ?

Brian Schwarz Brian.Schwarz at veritas.com
Fri May 30 12:26:02 UTC 2003


Although I can't help you with the specifics of LVM, I can tell you that
this is a common config for many companies. Many people use RAID5 in the
disk array because it is efficient, and offloads the parity calculations
from your server. Using host based RAID 1 will protect your application if
any of the disk arrays crash/fail. Multi-pathing will protect against HBA
failures as well as some switch or disk array port failures. I would
recommend keeping the SAN switches in two separate fabrics if possible,
eliminating a single point of failure in the FC switch mgmt software

If you ever want to take a look at a LVM alternative, the VERITAS Volume
Manager included in our Foundation Suite (we have a File System in there as
well) is very robust in these types of environments. More info at



-----Original Message-----
From: Rainer Krienke [mailto:krienke at uni-koblenz.de]
Sent: Friday, May 30, 2003 2:10 AM
To: linux-lvm at sistina.com
Subject: [linux-lvm] lvm in linux SAN enviroment ?


I recently posted about a problem with lvm (subject: Strange lvm on raid1 on

top of multipath problem posted on 05/29/2003). Well in between  I could 
image what the reason might be but I'm quite unsure if my theory is correct:

There are three machines and three hardware raid 5 devices connected by two 
fibrechannel switches to each other, so each host can see each physical 
partiton on each of the three disk devices.

Now on each host I configured *one* md-raid1 device of cource based on 
different physical disks. There is one more layer of abstraction because the

raid1 device is actually built upon a multipath md-device which uses 
different paths to the physical disks. This works just fine

Next I defined *one* physical volume on each of the three hosts on the md 
raid1 device and then defined a volumegroup consisting of one physical
on each of the hosts. Finally I created several logical volumes. So on each 
host there is exactly one raid1 device used as physical lvm volume and one 
volumegroup with several volumes defined.

The question is if such a setup can work or is it bound to fail? 

Each host can because of the san network see each existings *physical* disk 
which might be a problem for vgscan, but on the other hand I created the 
*physical volumes on  raid1 md devices* and these devices are different on 
each host and each host has exactly one such md device defined (it  cannot 
"see" the other md devices on the other hosts allthough it can see the 
underlying physical partitions).

Can anyone please comment on this scenario? 
If this setup is simply "wrong", how could I tell vgscan which might to my 
knowledge be the problem, that it should not scan all physical devices it 
finds to avoid trouble.

Thanks in advance 
Rainer Krienke, Universitaet Koblenz, Rechenzentrum
Universitaetsstrasse 1, 56070 Koblenz, Tel: +49 261287 -1312, Fax: -1001312
Mail: krienke at uni-koblenz.de, Web: http://www.uni-koblenz.de/~krienke
Get my public PGP key: http://www.uni-koblenz.de/~krienke/mypgp.html

linux-lvm mailing list
linux-lvm at sistina.com
read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/

More information about the linux-lvm mailing list