面向边缘计算的Storm边缘节点调度优化方法

边缘计算有高实时性和大数据交互处理的需求,边缘异构节点间的调度时耗长、通信时延高以及负载不均衡是影响边缘计算性能的核心问题,传统的云计算平台难以满足新的要求。文中研究了在边缘计算环境下Storm边缘节点的调度优化方法,建立了面向边缘计算的Storm任务卸载调度模型。针对拓扑任务在边缘异构节点间的实时动态分配问题,提出了一种启发式动态规划算法(Inspire Dynamic Programming,IDP),通过改变Storm的Task实例的排序分配方式以及Task实例和Slot任务槽的映射关系实现全局的优化调度;同时,针对拓扑任务的并发度受限于JVM栈深度的缺陷,提出了一种基于蝙蝠算法的调度策略。实验结果表明,与Storm调度算法相比,所提算法在边缘节点CPU利用率指标上平均提升了约60%,在集群的吞吐量指标上平均提升了约8.2%,因此能够满足边缘节点之间的高实时性处理要求。

Edge Computing-oriented Storm Edge Node Scheduling Optimization Method

Edge computing has the demands of high real-time and big data interactive processing.The key problems of edge computing performance are long scheduling,high communication latency and unbalanced load among the edge heterogeneous nodes.Traditional cloud computing platforms are difficult to meet these new requirements.This paper focuses on the scheduling optimization method of Storm edge nodes in the edge computing environment.Firstly,a Storm task offloads scheduling model for edge computing is established.And then a heuristic dynamic programming algorithm is put forward to realize real-time dynamic allocation of topological tasks among edge heterogeneous nodes.By changing the sorting method of the Task instance and the mapping relationship between the Task instance and the Slot,the global optimization scheduling is achieved.Aiming at the problem that the concurrency of topological tasks may be greater than the maximum depth of the JVM stack,a scheduling strategy based on bat algorithm is put forward,the global scheduling scheme is calculated according to the information of the topology task and the CPU information of the edge node.Experiments show that compared with the current Storm scheduling algorithm,the proposed algorithm has an average increase of about 60%in the CPU utilization metrics of the edge node and an average increase of about 8.2%in the throughput metrics of the cluster.Therefore,the proposed algorithm can meet the high real-time processing requirements between edge nodes better.