Not all IOPS are created equally

By Ryan VerVaecke, Sr. Principal Architect –

Performance metrics play a significant role in the operation of the modern data center. We need to know the speed at which different processes function and are available. So, IT focuses heavily on a number of analytics derived from performance data to assess the health of its systems. There’s a problem, though, with assuming that every metric offers the level of insight needed to make decisions. For example, we frequently see service-level agreements (SLAs) from different vendors based on input/output (I/O) operations per second, or IOPS. I/Os are a pretty common gauge of performance, and it’s common knowledge in this industry that the more requests a solution can process, the better it’s likely to perform. Not every data center component’s I/Os are created equally, though. Some consume more resources, and it’s just not practical to assume that because a solution delivers strong IOPS for one process that it can do the same for another.

Milliseconds count in performance

This an especially crucial consideration in fields, such as banking and finance, where even an added millisecond can hurt performance. Banks need to process countless transactions almost instantly. Hospitals can’t wait for patient records. Assessing performance based on IOPS without context related to the kinds of data being transmitted doesn’t really provide the guaranteed availability many SLAs include. The transactions will eventually be processed, but speed is a critical consideration for so many industries. If these processes aren’t completed quickly, they delay operations. For financial services companies or hospitals, additional latency is just unacceptable.

Understand what workloads matter for you

For any CIO or other IT decision-maker, the I/Os most vital to your company’s operation are going to be different from those in another organization. Even in the same industry, the processes are going to vary slightly. When it’s time to make new decisions on infrastructure components, consider any and all differences from one company to yours. When a solution provider includes IOPS in SLAs, the kinds of processes with which their machines reached those figures may be much less rigorous than yours. Or, they may have come from a smaller, less complex infrastructure not as susceptible to latency issues. These SLAs are often based on availability, rather than performance, and IOPS without context are a potentially misleading metric.

There are other similarly misleading performance metrics. Like any other statistic, though, analytics don’t deliver any value if they don’t come within context. CIOs, especially those charged with managing technology for hospitals, banks and other sensitive industries, need to understand that infrastructure performance management isn’t about more than just availability. It’s about optimization and efficiency across the board.

Want to know which metrics are most significant for your infrastructure performance? A VirtualWisdom4 demo will put you on the way to finding them.