Rather than simply throwing large amounts of on-site generation (eg photovoltaic panels) at the problem, HP's approach is to apply some smarts to balance supply and demand, for example by running non-realtime workloads during daylight hours to take advantage of solar power.
The process starts with a prediction module that forecasts workload demand and the cost and availability of resources, including power. That feeds into a planning module that optimises the workloads according to operational and strategic goals.
The execution module carries out that plan, managing workloads and energy consumption in real time, while the verification and reporting module identifies and remediates differences between the plan and its execution.
The test bed for these ideas was the sustainable data centre at HP Labs in Palo Alto. The company plans to put the architecture into a production setting at its Fort Collins (Colorado) research facility, with further development as part of its Moonshot (energy-efficient computing) project.