Scaling Supercomputing with Stranded Power: Costs and Capabilities

Fan Yang; Andrew A Chien. 11 February, 2016.
Communicated by Andrew Chien.


Power consumption (supply, heat, cost) and associated carbon emissions (environmental impact) have become critical challenges in scaling supercomputing. Recent work pro- posed exploiting stranded power, renewable power generated at times/places that render it useless to the power grid, for scaling supercomputers in the face of these challenges [1]. These studies showed that stranded power can be used to expand the capabilities of supercomputing systems. This stranded-power- based extension is named Zero-Carbon Cloud (ZCCloud).

We build on these results with deeper analysis of the phenomenon of stranded power, characterizing its temporal, geographic, and interval properties. Combining the resulting stranded power characteristics with production supercomputing workloads and a data center total-cost-of-ownership (TCO) model, we explore capabilities and costs of scaling supercomputing based on stranded power. Specific dimensions of exploration include power price, computing hardware price and power density. ZCCloud resources can reduces the cost of extension by 50%, and be 35% more cost-effective. With higher power price, cheaper computing hardware and higher system power density, these benefits can grow to 50%, 105% and 118% respectively.

Original Document

The original document is available in PDF (uploaded 11 February, 2016 by Andrew Chien).