基于可靠性的多核系统硬实时任务并行调度

Reliability-based Parallel Scheduling of Hard Real-time Tasks for Multicore Systems

提出了一种新的基于可靠性的多核系统硬实时任务调度方法,并给出定量可靠性模型。首先,每个应用程序都由任务组成,并采用有向无环图建模,显示任务执行周期、任务依赖性及其通信量。其次,提出聚类多数投票方法,在纯大多数投票并行性和纯再执行序列化间进行权衡。聚类多数投票方法是具有确定性定时行为的半并行方法,同时也减少了通信量。然后,提出用于硬实时应用的任务映射策略,试图最小化通信开销和任务图的副本实例,同时保留预定义的可靠性阈值,实现更高的算法性能。仿真结果表明,该方法具有较大的片上网络尺寸、较高的可靠性阈值和接受率,提高了算法的效率。

In this paper,a new hard real-time task scheduling method based on reliability for multicore systems is proposed,and a quantitative reliability model is given.Firstly,each application is composed of tasks and modeled by directed acyclic graph to show the task execution cycle,task dependency and traffic.Secondly,a clustering majority voting method is proposed to balance the parallelism of pure majority voting and pure reexecution serialization.The clustering majority voting method is a semi-parallel method with deterministic timing behavior and reduces the traffic.Then,a task mapping strategy for hard real-time applications is proposed,trying to minimize communication overhead and copy instances of task graphs,while retaining predefined reliability thresholds to achieve higher algorithm performance.The simulation results show that this method has larger network on chip size,higher reliability threshold and acceptance rate,and improves the efficiency of the algorithm.