کنفرانس مهندسی برق ایران

صفحه اصلی / بیست و نهمین کنفرانس مهندسی برق ایران

Improving the Reliability of Multicore Embedded Systems through an Evolutionary-based Task Scheduling Approach

نویسندگان :

Athena Abdi¹ Hamid R Zarandi²

1- K.N.Toosi University of Technology 2- Amirkabir University of Technology

کلمات کلیدی :

Multiprocessor embedded systems, Reliability, Power Consumption, Multi-objective optimization ,Evolutionary algorithm, Task scheduling.

چکیده :

In this paper, a task scheduling approach based on Strength Pareto Evolutionary algorithm (SPEA2) to improve the reliability of multi-core embedded systems is presented. Multi-core embedded systems are widely employed in many applications such as transportation, health-care, electricity grid and so on. Reliability is one of the major concerns of these systems due to its effect on lifetime of processors. To efficiently enhance reliability, its mutual impact on other design parameters such as performance and power consumption should be considered. Moreover, reliability in terms of soft and hard error rate is very important due to employing the embedded systems in safety-critical applications. The mentioned criteria are not independent and their joint optimization is associated with some complicated trade-offs. Considering these criteria at the system-level and optimize them during the task scheduling process, is one of the most effective methods. In this paper, we employ SPEA2 as a powerful evolutionary algorithm for solving the defined multi-objective optimization problem during task scheduling process. Thus, the order and place of executing each task of application are determined during the optimization process of the mentioned design criteria and based on the precedence limits. To demonstrate the effectiveness of our proposed method, several experiments are performed that check the ability of our approach in exploring the design space and meeting the existing trade-offs among the considered criteria during enhancing the processing system reliability. Moreover, the proposed method is compared to related studies in terms of design space exploration and joint optimization capability. Based on these experiments, our proposed method outperforms related studies and provides a more proper solution set based on the joint optimization of reliability, power consumption and performance.