MSCS quorum disk best practice ?

[hraju]

Hello,

I was thinking about High availability of solution with MSCS quorum device. Best practice is to have at least 3 mirrored disks quorum. But if
- two of quorum disks reside on the same controller. The controller is single point of failure of cluster.
- two of quorum disks reside on the same location. The entire location failure will fail also the quorum and the cluster.
Therefor the best would be to have 3 mirrored disks, each on different location (and of course on different controllers).


Documentation says only this:
A quorum resource failure can be avoided by using a dynamic mirrored quorum
resource instead of a physical disk quorum resource. Unlike a physical disk quorum
resource, which contains a single disk, a dynamic mirrored quorum resource will
provide a high level of redundancy by allowing the mirroring of the quorum disk.
It is strongly recommended that a dynamic mirrored quorum contain three disks
because a cluster disk resource cannot be brought online unless a majority of disks
are available. With the quorum volume in a two-disk group, loss of one disk will
prevent the quorum volume from coming online and make the cluster unavailable.

Some other best practices I found:
http://www.techdata.com/content/tdenterprise/whitepapers/Veritas_vm_best_practices_wp_3_18_02.pdf

[Michael_Black]

Yes. You can mirror a volume across luns from different controllers or even different arrays as long as the luns/disks are in the same dynamic diskgroup.

[hraju]

Thank you for response.

And is there any advantage of using two disks that are on same controller ? Except of failure of one of the disks - but that can be accomplished by using raid.

[Wally_Heim · Accepted solution]

Hi Hraju,

Most customers get around the single controller (HBA) as being a single point of failure by using DMP.  With the MPIO based DMP you can have  up to 16 paths to each disk (lun).  

Some customers have numerous arrays in their data center so they can allocate luns from 3 arrays easily and without any new cabling.  Most are also using hardware RAID (1, 5 or 10) on the array side in addition to the software mirroring that SFW can do. 

SFW can have 32 mirror plexes per volume so 3 is just a recommendation.  You can add much more if you need to. 

There are a lot of things that can be done to over come failure points.  The issue is what do you and your company see as the biggest concerns, how much do you want to spend to cover these concerns and what is an acceptible level of risk this application at your company?

Thanks,
Wally