Originally posted by: TTF
Here's the scenario:
A single Unix host with 2 HBA cards (each with a single FC port).
2 Fibre switches (not connected) i.e. 2 Fabrics
One AMS1000 array with 0A 0C 1D & 1B on fabric 1 and 0B 0D 1A & 1C on fabric 2.
Over the coming months, the Unix host will be assigned numerous LUNs with varying levels of I/O.
For each LUN configured, what is the best practice for the number of controller ports to use?
(Obviously I intend to use multi-pathing software on the host.)
My experience is with EMC hardware where on a given array all ports are pooled for all hosts. So for an array with 8 ports, 4 will be used by the controller that owns the LUN and the other 4 will be there in standby should the first controller fail. So I'd see balanced I/O across the 4 active ports and no I/O across the 4 standby ports.
I have seen HDS arrays configured with just 2 ports (one on each controller) for each LUN and an attempt made to manually balance I/O across all the ports for all the LUNs.
In a scenario where it is nigh on impossible to balance the I/O manually across the 8 ports, surely it is better to pool the resources or is there a good reason not to do this?