针对嵌入式多处理器平台对节能和散热控制的实际需要，本文提出了一种以减少计算资源占用为目标的实时程序划分方法。该划分在保证实时程序可调度性的基础上，将程序划分为多个任务子集，并以资源预留的形式为各子集分配计算资源。划分算法首先以缩短关键路径长度为目标，以最大程度满足程序的可调度性。之后，算法将尽可能多的任务放置到同一子集中以减少总的有效带宽，从而节省所需资源。同时，通过设置任务的截止期与激活时间，任务间次序约束及数据传输时延被转化为任务的时间参数，使得各任务子集可作为相对独立的模块进行分析与调度。实验结果表明，相比于传统以缩短总调度时长为目标的划分算法，本文方法得到的划分结果平均节省资源占用15～30％。%Regarding the ever-increasing demand for reduction of energy consumption and heat dissipation in embedded sys-tems ,a partition method is proposed to reduce the computational resource requirement for real -time applications .The method parti-tions a real-time application to multiple task subsets without violating its schedulability ,and assigns computational resource to these subsets in the form of resource reservation .To guarantee the schedulability ,it first tries to reduce the length of the dynamic critical path of the application ,and then packs tasks as much as possible to reduce the required total effective bandwidth .Additionally ,the method transforms precedence relations and communication latencies between tasks to real-time attributes using deadline and activa-tion time assignment ,which increases the modularity of each task subset .The results of extensive simulations show that ,compared with existing partition methods that aim to reduce makespan ,the proposed method achieves 15 ~ 30% less resource requirement in average case .
展开▼
