2013-01-10 01:15 AM
In a Fabric Metro Cluster does the root aggregate have to be sync mirrored to the remote site?
I think the answer is no.
Also do Aggregates for site A have to be on different physical shelf to the mirrored aggregates from Site B due to hardware ownership?
2013-01-10 02:23 AM
Yes, root aggregate must be mirrored, otherwise no takeover can happen if site is lost.
You are correct, with hardware based disk ownership each filer owns whole shelf.
2013-01-10 11:28 AM
You don't HAVE to mirror the root aggregate but my advise to MetroCluster customer that does not want to mirror everything is - "If you only mirror one thing, that should be the root aggregate". As aborzenkov says a disaster takeover will not be possible and then you have pretty much lost the purpose of running a MetroCluster.
Modern NetApp controllers use software disk ownership, in standard setups as well as in MetroClusters.
The last controllers that used hardware disk ownership were the FAS3020's and FAS3050's.
With software disk ownership, you can mix ownership in individual shelves.
My advise is to refrain from doing so if possible. Using whole shelves and dedicated switch ports (Fabric MC) or HBA ports (Stretch MC) for disks owned by the 2 controllers will make day-to-day operations easier. You can have disk.auto_assign enabled and working for you when you replace disks, identifying a disk to be replaced is easier etc.