Synchronization-Aware Dynamic Thread Scheduling for Improving Performance and Saving Energy in Multi-core Embedded Systems

摘要

Nowadays, chip multi-processors (CMP) are being widely used in embedded systems due to their superior performance for parallel computing. However, they also incur significantly large power consumption. To solve this issue, designers of embedded operating system must provide an efficient thread scheduling algorithm, which not only maximizes the system performance, but also minimizes the energy consumption. Further, if the scheduler makes decisions without considering the precedence relationships among threads, the decisions could conflict with the thread behavior which could result in poor performance and large energy consumption. In this paper, we propose a synchronization-aware dynamic thread scheduling algorithm (SA), which reduces the busy-waiting time caused by spinlock, and with performance improvement and energy saving. SA has two major objectives, including (1) overall high performance, in terms of less completion time and less turnaround time, and (2) low energy consumption for all threads. The experimental results show that SA indeed improves performance and reduces energy consumption compared to the original scheduling algorithm of Linux kernel version 2.6. In a real-world case, the Digital Video Recording (DVR) system, SA achieves performance speedup by at most 1.21 and saves the energy consumption by at most 28.6%.