This article can also be found in the Premium Editorial Download "Storage magazine: How to distance your data from disaster."
Download it now to read this article plus other related content.
Of classes and tiers
To most storage administrators the concept of tiered storage centers on establishing multiple hardware platforms. For example: tier one may be an enterprise array, tier two a midrange system with Fibre Channel drives, and the same midrange array with SATA drives may serve as tier three. Essentially, the approach is as follows:
This focus on hardware is a mistake. Users don't care what storage platform an application is running on, but they do care about the level of service they're receiving. I realize this isn't always the case in the real world, and that users often dictate their preferred platforms. But this is largely because they've been conditioned to think and speak "hardware" to their infrastructure teams. This restricts IT and obstructs the development of a business-aligned, cost-effective storage infrastructure. It's how we got into the "90% of data on tier one" mess in the first place.
Instead, the discussion should be based on an application's required data management attributes. The goal is to establish classes of service rather than tiers of storage--a subtle, but critical, distinction. It shifts the conversation away from hardware to business needs and alignment. As shown in the next chart, the focus is on defining business requirements that can then lead to an application or data classification framework. This, in turn, leads to the definition of service levels and the creation of a service catalog. Then the technology to support each service level is selected. These services may--or may not--require multiple tiers of storage. In many cases, service-level distinctions will be based on applying different features of a single hardware platform, for example, RAID 5 vs. RAID 10, frequency of split mirrors, or synchronous vs. asynchronous vs. no replication.
Building the classification framework is the key to moving toward an aligned infrastructure. Classification can be done at the application or data level, although it's most practical to start at the application level. There are several different dimensions of classification that must be considered. Each dimension becomes a classification category, as shown in "A data classification framework".
This was first published in May 2006