What Makes Storage Primary?

Over the years I've had many conversations with users about their primary, secondary and sometimes tertiary storage. Too often the conversation quickly moves to products and applications with the members of the fraternal order of steely eyed storage guys insisting that only a Symmetrix, USP-V or similar supporting OLTP applications could be considered primary. So I have to ask: What makes storage primary?

In the last century, it was pretty simple to tell the difference between enterprise- or server-grade storage hardware.  It had SCSI or Fibre Channel interfaces, inherent RAID, and it was optimized for reliability and OLTP performance. When SANs came around, we mirrored the RAID controllers and their caches, connecting our servers via Fibre Channel.

Sometimes the arbitrary rules and standards users set up resulted in buying decisions that are just plain weird. Years ago, a client of mine had a corporate standard that SANs and servers would always use SCSI or Fibre Channel disks. Along came a project that needed terabytes of file storage, and they went off and bought an Iomega NAS appliance. Since it was a NAS and not a server, it could use ATA disks without violating the standards. I should note that Google's study of disk failure rates didn't show a significant difference in failure rates for ATA drives as opposed to more traditional enterprise devices.

Many of you would say that primary storage was block-oriented, but even that definition breaks down today. In some industries like entertainment or oil and gas, the critical applications use file, not block, storage to render the special effects for the next superhero movie or to find the next offshore oil site.

To me the definition is simple. Primary storage is where data is created. Secondary storage is populated with copies of data that was created on primary storage. Primary storage should have the features required to protect its contents, and will hold the only copies of that data that exist. Those features start with RAID, object mirroring or some other method of protecting data against hardware failures like the erasure/dispersal codes used by NEC's HydraStor and CleverSafe, and they will have a regular backup. If the data is valuable enough, add snapshots and replication. If the design of the system meets the organization's requirements, SATA drives duct taped to motherboards could be primary data.  Not that I'd want to work in that data center, but as my friend Curtis Preston says, "I'm just saying."