Finding better metrics for evaluating data center performance
GigaOM’s Stacey Higginbotham reports on the need for better metrics for evaluating the performance of today’s data centers. I’ve written before about the shortcomings of power usage effectiveness (PUE) as the primary means of determining how energy efficient a data center is. Where I believe we’ll ultimately head is toward measuring application tasks and processing on a per watt level. Essentially, how much work did my data center do per watt of energy expended.
Others, like GigaOM research analyst Dave O’Hara, believe that we also need to be looking at operational expenditures and capital expenditures via the prism of energy. O’Hara asks these questions this way:
- Capital expenditures: How much does this data center cost to build, per megawatt?
- Operational expenditures: How much does it it cost me to run the data center, per megawatt?
Defining costs per megawatt is how data center real estate functions. When a company wants to find space to build a data center, it assesses the available power for that facility and then prices the real estate at a per megawatt cost. There’s certainly value to defining the performance of a data center in how much its costing you for the amount of power it uses as well as the cost to build it. You can get a real sense of how sophisticated and redundant that data center is and then match the mission critical tasks up with the more costly, and presumably faster and more reliable, data centers. While less critical tasks can be farmed out to cheaper per megawatt data centers. We might even see a spot market where data center processing time could be priced at different rated tiers of service, the way you pay different amounts for bonds that have different ratings and carry different risks.
What still needs to happen is performance per watt metrics to tell users how efficient the actual hardware in the data center is operating. Yes, that metric is baked into cost per megawatt of operation but it’s a step further and matters to companies like Calxeda and Tilera, which are trying to figure out how to optimize server performance at low power, or companies like Power Assure, which are trying to help data centers better manage their energy use.