multipath disk failover?

[Ian Dennison_1 · Accepted solution]

You can have up to 8 paths for the same disk (the primary plus 7 alternates) using LVM.

Use the vgextend command and LVM will figure out that it is the same disk that is already present.

When one of the paths fails, you will see an error in syslog and/or dmesg, but the Disk will fail over to the alternate paths.

Is that what you were after?
Share and Enjoy! Ian

[Chris Cruz_1]

Is there a way to do multipath disk failover?

Somebody told me of this scenario:
Two fibre cards that will have different cxtxdx values.
The first fibre card disk contains the data in a logical volume and if you add another drive from the second fibre card then the first is primary and the second disk becomes the failover.
Fibre card one - use the command line or sam to create vg10 with disk c0t0d0
with the logical volume called work mounted at /work.
Fibre Card two - add the fibre card's c1t1d0 to vg10.
vgextend /dev/vg10 /dev/disk/c1t1d0
I'm told that because c1t1d0 doesn't have a logical volume or file system on
it, then this second drive address via the fibre card automatically becomes the
secondary and will be active when fibre card one dies.

Is this true?

[Ian Dennison_1]

In addition, you can balance the load for multiple disks by making one Fibre Card the primary link, then the next disk has the other Fibre Card as the Primary Link.

There are some tools available from Veritas that allow both Paths to the disk to be used at the same time.

Share and Enjoy! Ian

[Vincent Fleming]

Chris,

You are correct.

You do not say if you have an array or JBOD, but there are other options as well, depending on your storage device. If it's an HP storage array (a current model) you can use the AutoPath or SecurePath software (different package, depending on the which storage array you have) to load balance as well as provide path fault tolerance.

Good luck!

Vince

[Sridhar Bhaskarla]

Hi Chris

//

vgextend /dev/vg10 /dev/disk/c1t1d0
I'm told that because c1t1d0 doesn't have a logical volume or file system on
it, then this second drive address via the fibre card automatically becomes the

//

That was incorrect. c1t1d0 is nothing but c0t1d0 except that it is accessed through a different link.

That might be a typo. c1t1d0 cannot be an alternate link to c0t0d0. Only the c? number changes for the secondary link.

-Sri

[Eugeny Brychkov]

Great replies. Sridhar, usually, in modern enviroment c? only changes, but if you have Nike disk array or autoraid (SCSI connected devices), then controllers can have different SCSI Ids. Example: connect autoraid to one HBA (daisy-chaining controller X and Y), controller X has Id1 and controller Y has Id5. Then two paths for LUN3 will be
c1t1d3
c1t5d3
Eugeny