Energy-efficient workload mapping in heterogeneous systems with multiple types of resources

Given the need to achieve higher performance without driving up power consumption, energy usage and heat dissipation, most chip manufacturers have shifted to multicore architectures. An important subcategory of such architectures are those that are heterogeneous in design. By providing heterogeneous processing elements with different performance and energy characteristics in the same system, heterogeneous architectures are expected to provide more flexibility for better energy efficiency. Indeed, prior work has shown that heterogeneous systems are able to reduce power and energy consumption compared to homogeneous systems. However, without power-aware application and data mapping methods that explicitly consider heterogeneity, such advantages cannot be fully exploited. Such methods are of interest for other reasons as well. For example, excessive power consumption may cause system unreliability (due to power constraints and local hot spots).