Performance Improvements Using Application Hints on a Multicore Embedded System

Multicore processors are increasingly adopted to embedded systems like smartphones and tablets as user applications on such devices become more complex and require high performance. However, it is in doubt that the user applications for embedded systems with multi-processing capability exploit the power of the multicore CPU fully. Unlike servers or desktop PCs, power-performance balance is most important in embedded systems. Thus if an application is not carefully designed to efficiently use the multicore CPU, or the system is not aware of it, the use of multicore might result in unexpected failure, such as little performance improvement with high power consumption. In this paper, we present a framework for efficient use of multicore CPU in embedded systems. The proposed framework monitors the usage of the computing resources such as CPU cores, memory, network, and the number of threads. Then it manages the number of CPU cores to be assigned to the application using the resource usage hints. We have tested the framework using SunSpider benchmark with FireFox and Midori Web browsers on an embedded system with Exynos4412 quad-core. Experimental results show that by managing the core assignment and frequency scaling, we can improve the energy efficiency along with the performance.