Network RAID over dissimilar Cluster set

[Roberto La Verde]

Hello all,
I'd like to know if it possible (and supported) to have a Network RAID 2 LUN in a scenario where there is one SAS Starter kit 4.8 TB physically placed on a primary site and another SATA Starter kit 6 TB on a secondary (DR) site.
The two sites are linked toghether by a Gbit trunk (for both iSCSI and LAN) and the LAN network is unique (same subnet).
I'd like to place the two SAS heads in the same physical place and the same for SATA ones.
Is it possible?

[Gauche · Accepted solution]

Possible and supported... Yes.
You can mix any node types in the same cluster.

Suggested... usually not.

Since data is stipped and mirrored equally across all nodes, and writes are not acknoledged until all mirrors receive it (otherwise its not syncronous), the slower SATA nodes will slow down the faster SAS nodes in the cluster. Exactly how good that system would perform is unpredicatble, both because testing mixed clusters for performance in not normal, and the speed / latency of the inter site link has a huge affect.

If you want to do something similar and not let the SATA slow down the SAS nodes, you can create two clusters and setup remote copy replication between them. Since this is not syncronous replication and they are not the same cluster the SATA will have no affect on the SAS performance. That might be of no help to you though as it sounds like you were after syncronous replication.

Hope that helps.

[Roberto La Verde]

Many thanks, Gauche.
By the performance point of view, it sounds flawless. I was looking for a cost effective BC solution, considering also that it could be coupled with FT technology from VMware Ent+.
Do you have any idea about the difference in performance (pure IOPS) between SAS and SATA boxes, even using hardware RAID 1+0 on the SATA ones?
Your alternative solution sounds good and applicable for non-mission critical VMs and will be a practical option.
But, what about consistency?

[Bryan McMullan]

Roberto, you would probably need three nodes to do what you are looking for. Two nodes at your primary site and one node at your remote site. You could do Network Raid 2 at the primary site to protect against node failures and remote snapshots to the secondary site to protect against site failures.

In our usage scenerios, the SATA units are 15-30% slower than the SAS drives. But this is highly dependant on the internal raid setup of your nodes. (We use RAID 1+0 for production clusters and JBOD or RAID 5 for standby/DR units.)