As an extension of cloud computing, the edge computing has become an important pattern to deal with novel service scenarios of Internet of Everything (IoE) under 5G, especially for the delay sensitive computing tasks generated from edge equipment. The edge computing provides the key support to meet the characteristics of delay sensitivity by deploying servers near network edges. However, a great many uneven distributed computing tasks in different network edges usually lead to task processing delay bottleneck for single Edge Computing Server (ECS). Tasks assignment is mainly based on the local ECS status without the global network view considered, which also easily leads to unbalanced task loads among multiple ECSs. In this paper, the novel networking idea of Software Defined Network (SDN) is introduced into the edge computing pattern. The logically highly centralized control plane consists of multiple physically distributed ECSs, so as to collaboratively assign computing tasks in a global view. In order to optimize the task assignment and minimize the task processing delay, three schemes are proposed in this paper. The scheme of assessing the ECS’s task computing features is firstly proposed, then the scheme of predicting the ECS’s future unit task processing time is presented. Thus, different types of computing tasks can be assigned to appropriate ECSs that are better at dealing with them with processing delay minimized. Furthermore, the scheme of optimizing the delay of task processing time estimation is devised, so as to further improve task assignment efficiency. Experimental results show that the proposed mechanism is able to optimize the task assignment and minimize the task processing delay more efficiently than the state of the art. Specifically, our mechanism is capable of improving the average unit task processing delay and the ECS load balancing degree by about 14% and 23% respectively, compared with corresponding work.

作为云计算的扩展，边缘计算已成为处理5G下万物互联（IoE）新颖服务场景的重要模式，特别是对于边缘设备生成的时延敏感计算任务而言。边缘计算通过在网络边缘附近部署服务器来提供关键支持，以满足延迟敏感性的特征。但是，不同网络边缘中的大量不均匀分布式计算任务通常会导致单个边缘计算服务器（ECS）的任务处理延迟瓶颈。任务分配主要基于本地ECS状态，未考虑全局网络视图，这也很容易导致多个ECS之间的任务负载不平衡。在本文中，将软件定义网络（SDN）的新颖网络思想引入边缘计算模式。逻辑上高度集中的控制平面由多个物理分布的ECS组成，以便在全局视图中协作分配计算任务。为了优化任务分配和最小化任务处理时延，本文提出了三种方案。首先提出了评估ECS任务计算特征的方案，然后提出了预测ECS未来单元任务处理时间的方案。因此，可以将不同类型的计算任务分配给更擅长以最小化处理延迟来处理它们的适当ECS。此外，设计了优化任务处理时间估计延迟的方案，以进一步提高任务分配效率。实验结果表明，所提出的机制能够比现有技术更有效地优化任务分配并最大程度地减少任务处理延迟。具体而言，与相应的工作相比，我们的机制能够将平均单位任务处理延迟和ECS负载平衡度分别提高约14％和23％。