Why Tiering and QoS Matter in All-Flash Array Selection

It is an article dated January 4, 2018 by Ken Clipperton, managing analyst, DCIG, LLC.

Why Tiering and QoS Matter in All-Flash Array Selection

Many organizations are using all-flash arrays (AFAs) in their data centers today.

When asked about the benefits they have achieved, two benefits are almost always top of mind. The first benefit mentioned is the increase in application performance. Indeed, increased performance was the primary rationale for the purchase of the AFA. The second benefit came as an unexpected bonus; the decrease in time spent managing storage.

Based on these initial wins, organizations are seeking to extend these benefits across the entire application portfolio. Consolidating many applications onto each AFA can create resource contention, reducing the performance of other applications that share the array.

Sophisticated data tiering and QoS features mitigate the impact of resource contention, giving business critical applications a greater portion of the arrays performance resources. Thus, data tiering and QoS features enable organizations to accelerate more applications without reintroducing storage management overhead.

Why QoS Matters in an AFA
QoS features come into play when multiple applications share the same storage system. Just as thin provisioning virtually expands the capacity of a storage system, QoS virtually expands the performance of a storage system. It is like thin provisioning for performance.

AFAs implement QoS features in a variety of ways. Some implement static maximums or minimums in terms of latency, bandwidth and IO/s. Other AFAs implement QoS policies as dynamic attributes that give priority to workloads based on classifications such as High, Medium and Low.

Different applications have different storage requirements, and some applications are more important than others to an organization. QoS features that map to these differing application requirement, and business priorities add real value to an AFA.

Why Tiering Matters in an AFA
Tiering matters in an AFA because persistent storage media are proliferating based on characteristics including latency, bandwidth, durability and cost. A storage system that can manage data placement based on these characteristics can deliver optimal performance at the optimal cost.

How Many Tiers is Enough in an AFA?
Some AFAs provide a single tier of storage. Other provide two tiers: a performance tier, and a capacity tier. Other ones provide three or more tiers. I asked an experienced storage engineer how many tiers were required to optimize performance and cost. His answer, “Fourteen.” That number may raise some eyebrows, but when one takes the varied characteristics of storage media into account, it is not hard to get to fourteen data tiers.

Those performance characteristics include:
• Latency
• Bandwidth
• IO/s
• Durability
• Cost

Advances in NAND flash, storage class memories and the adoption of the NVMe protocol create significant new options based on each of these performance characteristics.

Accelerating All Applications Without Compromise
AFAs that provide sophisticated tiering and QoS features can take greater advantage of advances in storage technology. Organizations implementing these AFAs can accelerate all applications without compromising the benefits they achieved with their initial all-flash array implementations-faster application performance and reduced storage management overhead.

DCIG is refreshing its research on AFAs, and is taking these data tiering and QoS capabilities into account as it evaluates these products. DCIG expects to publish reports based on its updated AFA research beginning in the first quarter of 2018.