Power-aware computing for multi-agent systems

Agent-based modelling is becoming a common technique for studying complex phenomenon in diverse fields including sociology, economics and biology. This technique is assisted by the continuing exponential and pervasive growth in computing power. In recent years, engineering limits on processor speed have spurred focus on multi-core processing as a means of continuing this growth. In order to harness this computing power, however, careful concurrent programming is required to develop multi-threaded applications. Furthermore, the power consumed by systems has come under increasing scrutiny from the standpoints of thermal capacity, energy availability, and sustainability. We investigated two schemes for reducing simulation power demands using The Game of Life as a representative simulation: a) event-driven computation and b) dynamic frequency scaling in multi-threaded rate-limited applications. Both schemes were found to significantly improve energy efficiency. Both event-driven and parallelized computation are central to the low power usage of biological computation compared to silicon.