实时系统中的熵容错模型研究

Research on Entropy-based Fault-tolerant Model for Real-time Systems

针对硬实时系统中采用的N重冗余容错模式存在代价大和容错效率不高的问题,在硬实时分布并行系统中提出了一种熵容错模型,将数据/任务按给定的约束分割为惯性和非惯性两个子集,其中惯性分量的未来状态是可估值的,非惯性分量是不可预测的。在容错实现中,对惯性分量不做冗余,在故障引起的间断期则对其估值;对非惯性分量进行多重冗余,从而使系统在故障活动期内仍能提供可信服务。应用证明该模型在惯性分量权重较高的系统中能够以低于2的容错因子包容多个故障,改善容错效率。

The N redundant models have the problerns of high cost and low fault tolerant efficiency in traditional hard real-time systems.The paper presents an entropy-based fault tolerance model for parallel and distributed real-time systems.In the model,data/tasks are partitioned into inertial and non-inertial subsets：an inertial subset that is predictable,and another non-inertial subset that is unpredictable.In realization of the fault tolerance,only the non-inertial segments are redundantly reduplicated,the output of the interrupted inertial segments will be complemented with predicted parameters during the fault-active period,consequently the systems can provide dependable service during fault active duration.Practical applications with high inertial weight properties have proved that this model can tolerate several concurrent faults with lower fault-tolerance factor than 2 and achieve cost-effective fault tolerance.