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

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.