3Par Logical disk and controller node ownership

[Andongl]

Hi Experts,

I am attending 3par online training, need some help to understand the basic concept. In the training notes it's mentioned "the logical disks are bound to a controller node so virtual volumes are load balanced across all controller nodes. " I don't quite understand the statement, if a logical disk is bound to one controller, how the failover is handled during the failure, does it mean if one controller failed, all the associated logical disks also failed?

Much appreciated for your help!

Best Regards,

Andong

[Sheldon Smith]

---

@Andongl wrote:

Hi Experts,

... if a logical disk is bound to one controller, how the failover is handled during the failure, does it mean if one controller failed, all the associated logical disks also failed?

A Logical Disk (LD) is bound to one controller (under normal operation).

With the (current) exception of small numbers of SSDs, the Physical Disks (PD) are arbitrarily split between the nodes of a controller pair. One node "owns" the even-numbered PDs, the other node the odd. Either node can access all the PDs on its shelf, but to even the load on the nodes, they share the work. The "showpd" command will list all the physical disks. Under the A and B Ports columns, there is an asterisk (*) beside a N:S:P data port. That indicates which node accesses (owns) that PD under normal operation. The RAIDsets are created according to the CPG pattern rules with each node using chunklets from the PDs it "owns".

When a controller node of a pair is offline (not normal), the remaining node of the pair operates all the PDs on the pair's shelves. And thereby, all the LDs of the node pair.

For a system with a small number of SSDs, the concept of Express Layout is added. The asterisk shows up beside both data ports: Either node can access the chunklets on the PD. The LDs are still under the control of the node that created them, so while both controllers access the PDs, LDs are still owned by one controller node or the other.

[Dennis Handly]

> if a logical disk is bound to one controller, how the failover is handled during the failure,

If you use showld, the Own column will show you the owner node and up to 3 backup nodes (if you have 4 nodes).

> One node "owns" the even-numbered PDs, the other node the odd.

FYI: This will only happen if all of the slots are filled in when admitpd/admithw is done.  I.e. it is the slot that matters not the PD ID.