一种面向多物联代理在线应用的弹性资源调度算法

An elastic resource scheduling algorithm for online applications from multiple IoT agents

在电力物联网场景下,不同的边缘物联代理产生类型和结构均不同的数据,运行不同的任务,并且这些任务之间可能存在依赖关系,如何将这些任务调度到合适的物联管理平台的数据处理模块是一个颇具挑战性的问题。尽管对计算机集群上的相关调度研究已有很多,但是一般计算机集群上的调度策略并不适合直接迁移到电力物联网场景上。因此,探究由存在依赖关系的任务构成的应用在一个分布式的物联管理平台上的在线调度决策是有意义的。提出了一个兼顾应用处理用时和负载均衡的联合优化问题,并通过一个两阶段的启发式算法来求解。仿真实验结果表明:算法在较短的时间内达到了接近最优的表现,在大多数情况下,与理论最优结果的差距小于35%,因此能在较好满足处理时长要求的同时,兼顾物联管理平台的负载均衡。

In many Electric IoT scenarios,different IoT agents produce data with different types and structures and run heterogeneous tasks;these tasks may depend on each other.How to schedule these tasks to multiple data processing modules in the IoT management platform is a challenging problem.While many prior studies consider the scheduling decision on a computer cluster,it’s not appropriate to transfer scheduling strategies on a general computer cluster to our scenario directly.Hence,it’s meaningful to study the online scheduling for an application consisting of dependent tasks on a distributed IoT management platform.This paper proposes a problem that jointly optimizes for the processing time and load balancing,and design a two-stage heuristic algorithm.Simulation results shows that this algorithm can approach to the optimum in a short time,can thus meet the completion time demands as well as load balancing requirements.