Midrange Best Practice
LUN Assignment and Controller Ownership
9500V/AMS/WMS system controllers manage LUNs by ownership. This creates a reference for the controller's cache to manage hit/miss determination and for duplexed writes. Since management of a LUN starts with the parity group, all LUNs within a parity group should be assigned to the same controller to minimize controller-to-controller communication overhead.
Guidelines to Improve SATA Reliability
- One spare drive allocation per same number of RKAAT trays.
- FC drives cannot use SATA spare
- FC must use FC spare
- SATA must use SATA spare
- 6D+1P or less for Raid5
Zoning Guidelines
If you plan to connect different types of servers to the DF subsystem via the same fabric switch, you must use the zoning function of the fabric switch
Single-HBA zoning best simulates a single-initiator SCSI environment, in that HBAs do not log in to other HBAs, and no link-level services occur between adapters. This reduces the reliability concerns that can be raised when mixing driver revision levels, HBA types/vendors, and heterogeneous servers on the same fabric.
Configure each HBA with its own zone. Members of the zone should consist of the HBA and one or more Storage ports with the LUNs that will be used by the HBA. When dual paths are used, each path should have its own HBA connected to a separate controller that accesses the same Luns.
Alternate Pathing
The user should plan for alternate pathing to ensure the highest data availability. In general, 2 paths per LUN is enough - one path per CTL. There is normally little, if any, performance gain by assigning a LUN to 4 paths. Note that 9500V and AMS/WMS are Active/Passive subsystems. You must setup multipathing software correctly and never allow the Host to run Active/Active across controllers. HDLM will do this automatically. Solaris MPXIO needs special parameters to function correctly. Consult your Hitachi Account Team to obtain the proper procedures.
Queue Depth Setting
DF subsystems support 512 commands per Fibre Channel port. Excessive queue depth per LUN may cause error recovery routines to fail. For best performance, use the formula:
queue depth = 512 / (number of LUNs per port)
In general, set Queue Depth per LUN to be 8 for Windows, Solaris and HP-UX. Only increase if absolutely required.