2010-06-10 12:24 AM
I have a questions concerning a cluster environment including two FAS2020 with dual controllers.
Is it possilbe to assign a disk to more than one controller?
I want both of the controller to be able to access the disk for full rendundancy purpose.
The cluster will be in active-active running syncMirror.
How do I let both the controllers access the same disks?
Solved! SEE THE SOLUTION
2010-06-10 01:43 AM
I read something about setting up the system with a local and partner node.
Would it be possible then to have the two controllers on both of the ontaps in the cluster setup as local and partner?
2010-06-10 01:43 AM
So only one controller can own a disk at any one time. However, in a failover situation, where 1 node takes over the workload of the other, then all of the disks get taken over too and are presented through the second node.
From the way you are describing your setup it sounds like you want to have both nodes serve the same data. This isn't possible in ONTAP 7, but ONTAP 8 cluster mode does give you these options I believe. Its called an HP (High Performance) volume, if memory serves, and is pretty much as you describe. Each node owns its own disk, but data is accessible from two nodes. This was at least true in GX. You would have to check on its status in ONTAP 8. But for ONTAP 7 or 7-mode, it's not possible.
2010-06-10 01:46 AM
NO isn't possible to assign disk to more than one controleur
But for complet redondancy, of conroler, the CFO, is the nice solution, for redondancy of Controler and disk, the Active-Active conf. with Sync mirror is the real good solution , but if you want secure physicaly the system, you must have TOW controler even if isn't in Active-active, to made, snapmirror replication, or a little expensive solution, but the better => Metro-Cluster, with this you have tow controlers, in different room or building, and full sync-mirror beetwen us.
2010-06-10 01:50 AM
We are running Ontap 7.3.3.
So then it might not be possible to have the cluster setup the way I want.
We are planning to have just one raid group and aggregate on each of the nodes. The nodes will be in active-active mode using syncMirror.
Do you know what the best way is to have the second controller setup? It would be great to have it ready to be used if the first controller would fail.
2010-06-10 01:56 AM
Active / Active in this term means that each node is able to serve its own data. In most situations like the one you have described, I would assign very few (3) disks to one controller, and all of the rest to the second controller.
One the one with few disks simply have an aggr for vol0 and nothing else. It will essentially sit there idle, until a failure of failover occurs.
The other node will take the full load of the service. However when there is a failure, the full load will fail over to the redundant head. Essentially it would become Active / Passive. Not ideal, but the best way to handle this situation and guarentee performance during failure scenarios.
I'm sure there are more elaborate ways of achieving what would essentially be a parallel filesystem, but this would require additional software or devices to virtualise the front end. It also depends on the planned protocols you want to use. If it was CIFS for example, you could consider using DFS and have shares available on both. However, one would have to be a read-only mirror of the other (sync snapmirror between the heads) as DFS Sync doesn't work on anything but Windows hosts. If its SAN, there are some SAN Clustered Filesystems out there which could do the job. I've used a couple and they seem to do OK. I won't mention names here. NFS, there are options as well. Parallel NFS (pNFS) might be worth looking into, but it would involve some kind of cascading.
Hope this gives you some ideas whats possible and maybe something to start looking at.
2010-06-10 02:53 AM
Thank you for the answers.
So let's say I assign 3 disks to one of the controllers and the rest of the disks to the second one.
I set up the second one with an aggregate and one raid-dp group with one hot spare.
It sounds from your description that there should not be a raid-dp group on the first with only 3 disks, only an aggregate.
What would then happen in a failover situation?
If controller two, wich has the majority of the disks, would fail, would controller one with 3 disks jump in?
What would then happen with the 3 disks that it has?
2010-06-10 03:02 AM
You can have RAID_DP on the second even with three disks, 1 data, 2 parity, but the only question really is do you need that much protection of vol0. If you want it, its there. You could even go to 2 disks and just have RAID4. Again, it how efficient on storage you need vs protection levels.
In a failover situation the "personality" of the failed node would move the other node, disks and all. It would still appear to the outside world as two seperate devices, but everything would be managed from the same hardware in failover.
The disk ownership is attached to the "personality" which moves. Both filer heads have physical access to the disk, but are logically prevented from seeing those owned by the partner. So when the partner personality is relocated to other node, the disk ownership goes with it. Logically they are still two seperate services, just sharing hardware. Like VMs I guess (an anology I was struggling to avoid).
Does that help?