Making data center computations fast, but not so furious

Abstract

We propose an aggressive computational sprinting variant for data center environments. While most of previous work on computational sprinting focuses on maximizing the sprinting process while ensuring non-faulty conditions, we take advantage of the existing replication in data centers to push the system beyond its safety limits. In this paper we outline this vision, we survey existing techniques for achieving it, and we present some design ideas for future work in this area.