Model-Free Runtime Management of Concurrent Workloads for Energy-Efficient Many-Core Heterogeneous Systems

Modern embedded systems execute multiple applications, both sequentially and concurrently, on heterogeneous platforms. Determining the most energy-efficient system configuration (i.e. the number of parallel threads, their core allocations and operating frequencies) tailored for each kind of workload is extremely challenging. In this paper, we propose a novel runtime optimization approach with the aim of maximizing power-normalized performance considering dynamic workload variations. To reduce overhead and complexity, we adopt a model-free approach to runtime adaptation based on workload classification. This classification is supported by analysis of data collected from a comprehensive study investigating the tradeoffs between inter-application concurrency with performance and power under different system configurations. We conduct extensive experiments on an Odroid XU3 heterogeneous platform with synthetic and standard benchmark applications to develop the control policies and validate our approach. These experiments show that workload classification into CPU-intensive and memory-intensive types provides the foundation for scalable energy minimization with low complexity. Implementing this approach as a Linux runtime governor, we demonstrate that IPS/Watt can be improved by over 120% compared to existing approaches.