Emerging Technologies for Energy-Proportional Computation
Today’s servers are optimized for high energy efficiency only when used at peak or when at idle. Unfortunately, servers are often used at 30-50% capacity, in particular in large datacenters where medium loading is necessary in order to provide stability in the presence of workload bursts or machine failures. What the industry needs is energy-proportional computing, where energy efficiency remains high regardless of the load on a server. Microprocessors already have the technology to provide this energy proportionality through frequency and voltage scaling. What is needed is redesign of all other aspects of server system, from network hardware, to tertiary storage, to memory, to internal busses. Optical and capacitive coupled signaling technology promise to significant gains for both the inter- and intra-machine communication. Solid-state non-volatile storage offers an attractive alternative to mechanical disks.
UCSB has extensive research in CAD and circuit techniques to reduce energy consumption (Cheng, Marek-Sadowska), with funding from MARCO-DARPA and NSF. Finally, UCSB also has extensive research in architectures for emerging technologies (Chong, Sherwood), funded by NSF. Research in emerging technologies is heavily influenced by the available cooling technologies, as many devices have different characteristics at different temperature regimes. Furthermore, evaluating the combination of cooling and computing technologies requires an overall life-cycle analysis of the energy or carbon footprint of the manufacture, use, and disposal of both parts of the combination.