算力网络环境下基于势博弈的工作流任务卸载优化机制

Potential Game Based Workflow Task Offloading Optimization Mechanism in Computing Power Network

边缘计算虽然部分解决了任务上云导致的时延过长的问题,但由于通常只考虑端边云间的垂直协同,不可避免出现了“算力孤岛”效用,因而仍然难以满足工作流任务的低延迟执行需求.为了高效协同利用广域网上的算力资源,降低工作流任务的执行时间,亟需对算力网络中的工作流任务卸载和资源分配问题进行研究.首先描述了算力网络环境下面向多用户的工作流任务执行场景,并对该场景下的网络环境、工作流任务及其执行流程进行建模.其次根据优化目标建立工作流执行时延模型,以构建面向算力网络环境的多用户工作流任务卸载与资源分配问题.最后根据工作流应用的特点,针对链式工作流提出了一种基于势博弈的分布式工作流卸载算法.针对复杂DAG工作流提出一种基于动态资源权重的启发式工作流卸载算法.仿真实验表明,与其他算法相比,所提算法均能够协同广域网上的算力与网络资源,降低工作流任务的平均完成时间,从而有效提高了算力网络环境中的工作流任务的执行效率.

Although edge computing partially solves the problem of excessive latency caused by the tasks offloading to the cloud,it inevitably has the effect of“island of computing power”because of taking only vertical collaboration among the device,edge and cloud into consideration.Thus,it is difficult to meet the low-latency execution requirements of the workflow tasks.To efficiently and collaboratively utilize computing resources on the wide area network(WAN)to reduce the completion time of workflow tasks,it is urgent to study the offloading of workflow tasks and resource allocation problem in computing power network(CPN).Firstly,the multi-user-oriented workflow tasks'execution scenario in computing power network environment is described.Secondly,the network environment,workflow tasks and their execution procedures in this scenario are modeled.Thirdly,according to the optimization goal,a latency model is built to construct the multi-user-oriented workflow tasks'offloading and resource allocation problem.Finally,according to the characteristics of workflow application,a decentralized workload offloading algorighm based on potential game for chain workflow and a heuristic workload offloading algorithm based on dynamic resource weight for complex direct acyclic graph(DAG)workflow is proposed.Simulation results show that,compared with the other algorithms,the proposed algorithm can collaborate the computing resource and network resource in WAN,effectively reduce the average completion time of workflow tasks,thus effectively improving the execution efficiency of workflow tasks in computing power network environment.