CSR: Medium: Supporting Real-Time Data Flows on Heterogeneous Multicore Platforms

University Of North Carolina At Chapel Hill, Chapel Hill NC

Investigators

Abstract

This project addresses the challenges of implementing software upon heterogeneous multicore platforms, particularly in application domains where verified timing correctness is required. The multicore revolution is currently undergoing a wave of innovation in the form of heterogeneous processing elements -- processing elements that may differ with respect to functionality or processing speed. The presence of such heterogeneity means that choices must be made when allocating hardware resources to software components. The need to resolve such choices adds considerable complexity to resource allocation, and inhibits the adoption of such platforms by the embedded computing industry despite significant potential benefits in terms of balancing performance and energy usage requirements. This project will develop methodologies for implementing time-critical software upon heterogeneous platforms by systematically examining different models of heterogeneity, devising efficient real-time scheduling algorithms for each such model, and obtaining appropriate analysis techniques for each model that enable real-time constraints to be verified. Workloads will be modeled as data-flows that enable the representation of varying degrees of parallelism, with various possible combinations of constraints on preemptivity, inter-processor migration, et cetera, considered. This project both addresses fundamental scientific challenges in resource allocation theory, and seeks to develop appropriate software tools and methodologies to facilitate the adoption of results and insights that are obtained. Such results, tools, and methodologies are expected to accelerate the adoption of advanced heterogenous multicore platforms by embedded systems industries.