Distributed Systems for Energy Efficiency

Datacenter energy consumption is growing quickly and will soon become the limiting factor for scaling our computing power. My approach to this problem is massively distributed systems. For example, a big data workload may run on slow-but-energy-efficient hardware and compensate for the slowdown with scaling to a larger cluster.

Also, latency-critical applications (e.g., web services) have had to precompute for future big data queries which may or may not be used later. Flash bursts will allow transforming preprocessing for all potential queries into on-demand flash bursts, providing an opportunity for saving energy and other resources.

Lovelock explores building clusters out of SmartNICs: hosting services on wimpy-but-efficient SmartNIC cores can substantially reduce the carbon footprint of datacenter computing.