Run-time timing prediction for system reconfiguration on many-core embedded systems.
Journal of Systems Architecture(2019)
摘要
Many-core embedded systems usually have real-time constrains, which may work in hostile environment and operate continuously without supervision. However, system execution mode change and hardware malfunction could alter deployed applications’ response time and result in the violation of system’s real-time constraints. To accommodate such incidents, run-time system reconfiguration, which invokes dynamic application migration, needs to be supported on many-core embedded systems. As different migration strategies will impact system’s timing behaviors in different manners, it is vital to choose an appropriate one such that the system’s timing performance after the migration is still acceptable. The focus of this research is to predict system’s timing change induced by any migration strategy, which can be utilized to select the optimal migration strategy among all the possible choices. To be more specific, a two-stage timing prediction approach is investigated in this paper, where the offline stage is to train the initial model using historical data and the online stage is to fine tune the model at run-time. Extensive experiments have been conducted and the results validate the effectiveness of our proposed approach.
更多查看译文
关键词
Timing,Prediction,System reconfiguration,Real-Time,Many-core
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络