ONTAP Discussions

What is the purpose of Mirrored HA Pair?

N_ANDREEV
3,668 Views

NetApp has SyncMirror feature that lets you mirror two shelves for increased availability. SyncMirror saves you from a shelf failure. There is a MetroCluster which uses the same SynMirror feature beneath it and allows to fail over when the whole storage system fails.

And now we have Mirrored HA Pair which is SyncMirror again, but between two storage systems. Mirrored HA Pair doesn't automatically failover in case of storage system failure. Then what particular benefits does it give compared to SyncMirror and MetroCluster?

4 REPLIES 4

radek_kubka
3,668 Views

Well, I've read this thread as well:

https://communities.netapp.com/thread/25571

In all honesty I'd simply ignore the term "Mirrored HA Pair", as in real life you are dealing with two common scenarios for remote resiliency:

1) HA pair replicating (in most cases asynchronously) to a secondary system (either HA pair or single controller)

2) MetroCluster, where HA pair is stretched between two locations & SyncMirror is used to synchronously mirror the data

Maybe the author of the document wants to make a distinction when he talks about a corner case of scenario 1) when synchronous SnapMirror is used for replication from primary to secondary?

Regards,
Radek

N_ANDREEV
3,668 Views

Hi Radek,

Maybe the author of the document wants to make a distinction when he talks about a corner case of scenario 1) when synchronous SnapMirror is used for replication from primary to secondary?

Not really. Look what NetApp docs say:

Mirrored HA pairs provide high availability through failover, just as standard HA pairs do. Additionally, mirrored HA pairs maintain two complete copies of all mirrored data. These copies are called plexes and are continually and synchronously updated every time Data ONTAP writes to a mirrored aggregate. The plexes can be physically separated to protect against the loss of one set of disks or array LUNs.

radek_kubka
3,668 Views

Right, I stand corrected.

It looks like a description of local SyncMirroring, where controllers are not stretched. You do exactly the same plus stretch the controllers & get the well-known (& actually widely implemented in the field) MetroCluster - unless I'm missing something again!

N_ANDREEV
3,668 Views

Let me get this straight. You can use SyncMirror to avoid data loss in case of triple disk or storage shelf failure. HA pair saves you from controller failure. MetroCluster provides continuous data availability when a whole site goes down.

Mirrored HA Pair is HA + Inter-SyncMirror, but controllers aren't stretched. What are the odds of a whole storage system failure (controller + shelves)? If it's a multipath HA with Intra-SyncMirror, there are no single points of failure in this configuration.

Public