Skip to main content

Usage Centric Power Efficiency Metrics for Storage

In the course of the GAMES project and while developing SAMCEE, we considered the storage energy efficiency metrics that typically characterize the energy efficiency potential of storage devices and systems. These metrics do not characterize the actual use of the system in the data center. Consider a simple example of capacity energy efficiency of a 300GB HDD measured in GB/Watt. Clearly, a disk with a higher value of this metric provides more efficient raw storage than one with a lower value. Note that a large percentage (60% – 70%) of the energy consumed by this HDD is expended whether the entire or a small portion of the space is actually used by storing application data. This is due to the energy cost of spinning the disk. In one system configuration the space provided by this HDD may be fully used by applications, whereas in another configuration, only a small capacity is actually used out of the same HDD. The raw capacity efficiency metric calculated for the two cases is identical, but the energy cost of the capacity used by applications is very different.

Following this example, our rationale was that it is not enough to use storage resources that are potentially energy efficient: Optimizing the actual usage energy efficiency of the storage system operations is bound to result in improved energy savings. Targeting this issue, in our paper "Usage Centric Green Performance Indicators" (see also our paper and poster), we identified three usage centric energy efficiency metrics for storage. The three storage efficiency metrics work in a similar way to the corresponding metrics currently used by the industry, with the important difference that the metrics are computed on the basis of actual resource usage in a specific data center setting rather than the potential usage in a benchmark setting.

The three metrics are:

  • Capacity usage Metric (GB/Watt) – represents the current energy efficiency of storing the data of the user's applications. The metric is defined as the ratio between USED storage space by applications files (GB) and USED power (Watts).
  • I/O Throughput usage Metric (IOPS/Watt) – represents the current efficiency of the storage system while running the user applications workload. The metric is defined as the ratio between applications I/O rate (IOPS) and power (Watts) at a given time.
  • Data Transfer Throughput usage Metric (MBPS/Watt) – represents the current efficiency of data transfer while running the user applications workload. The metric is defined as the ratio between applications data throughput rate (MB per Second) and power (Watts) at a given time.

We contributed this usage centric metrics approach to the Datacenter/Storage Efficiency task force of the Green Grid Consortium industrial body, and it will serve as part of its future recommendations. We also recommend using these usage centric metrics as a best practice for storage management.