Splitting 1 multi-site cluster into two clusters?

[rhinkamper]

I currently have one 12 node cluster in two sites running network RAID 10 two-way mirror. Just recently I discovered that HP does not reccomend more than 10 nodes in a cluster.

So my question is, can I split my 12 node multi-site cluster into two 6 node multi-site clusters with the network RAID 10 two-way mirror between the clusters?

Will my storage space be impacted?

Is there a document that shows how to do this?

[oikjn]

you really have a 12 node multi-site cluster?  Thats impressive.  What is your bandwidth and latency between sites?

A cluster provides a single logical pool of space that you can assign to luns.  If you want that as a single space you cannot split it into two clusters.

You can migrate LUNs between clusters if you want, but even though you can move the data on a live syste, you will have to have a quick downtime while you switch from the old target initiator to the new initiator after the migration completes.  You can read about that in the manual under the subject "cluster migration"  I wouldn't suggest you migrate across clusters often.

Short story is that unless you are ok with splitting your luns across into to storage pools, you just have to keep it as a single large cluster.  That said, usually you don't need a single pool that large.

[rhinkamper]

Yes, unfortunately we have a 12 node multisite cluster replicating over a 40GB fiber link. I see what you are saying, and it looks like I should leave it as is. When we buy more storage the next go around I will just create a new cluster from there for our data to grow. 

[oikjn]

ah, someone who actually did it right.  Hard to come bye.

Depending on your current utilization and LUN spread and requirements I could see splitting out a main multi-cluster of 8 nodes and a side single site cluster of 2 nodes with remote replication to a 3rd 2node cluster.  Asuming you have some data that doesn't require complete sync across two sites and you can get away with async, that should reduce the load a bit on your main cluster and give all-around latency improvements.  It also keeps you from having all your preverbial eggs in one basket :)

That said, I believe the issue with >12 nodes is that you can run into the limits for iscsi connections depending on your MPIO/DSM configuration and LUN layout, but if you aren't running into that issue and don't have performance problems I wouldn't change anything.  "If it ain't broke- don't fix it" is generally a good thing to follow.