多核环境下基于图模型的实时规则调度方法

Method for Graph-based Real-time Rule Scheduling in Multi-core Environment

安全攸关反应式系统的核心要求是:必须在指定时间期限内完成对外部事件的检测和目标事件的响应,否则会产生灾难性的后果.随着安全攸关反应式系统对智能化需求的日益增加,将规则推理应用于这类系统成为必然趋势.规则调度是保证规则推理硬实时约束的关键.为此,提出了一种基于图模型的实时规则调度方法(graph-based real-time rule scheduling,简称GBRRS).该方法对基于事件图的实时规则推理过程进行建模,提出了基于图的端到端推理任务模型,并给出了端到端推理任务的调度算法,保证了规则调度的安全性.采用模拟实验对GBRRS方法进行了验证,实验结果表明,与DM-EDF方法(通过直接映射把规则上的推理操作转成推理任务后,用全局EDF算法对其进行调度的方法)相比,GBRRS方法在规则调度成功率上平均高出13%～15%,且在规则集的平均负载较高时,仍保持着80%以上的调度成功率.

Safety-critical systems detect external events, match the targeted event patterns, and give timely responding actions;otherwise catastrophic results will be incurred. With the increasing demand for intelligence in the safety-critical systems, applying rulebased reasoning to these systems has become an inevitable trend. Besides, rule scheduling is the key to assure hard real-time constraints within rule-based reasoning solutions. In this study, a solution to the multi-core rule scheduling problem, named GBRRS (graph-based real-time rule scheduling), was proposed. With the real-time rule reasoning process analyzed, how rules in safety-critical systems can be modeled as tasks using the graph mapping is described first, and the graph-based end-to-end reasoning task model, E2ERTG, is proposed. Then, a multi-core scheduling algorithm, GBRRS, is presented to guarantee each rule’s deadline via the control of the reasoning task’s deadline. Simulation- based experiments have been conducted to evaluate the performance of GBRRS. The result shows that GBRRS remains a rule success ratio above 80% even with relatively high workload of the rule set and is superior to DM-EDF by average 13%~15% in terms of rule success ratio.