A2C Learning for Tasks Segmentation with Cooperative Computing in Edge Computing Networks
Yukun Sun, Xing Zhang
Abstract
With the evolutionary development of computing-intensive and delay-insensitive applications, partial computing offloading in cooperative edge computing networks is considered as a promising technology to reduce tasks execution delay. However, existing researches focus on either splitting one task to several subtasks without exact proportion or splitting each of multiple tasks into hard two parts. In this paper, we consider splitting multiple computing-intensive tasks to several subtasks simultaneously. Accordingly, a joint tasks segmentation and parallel scheduling with cooperative computing problem is formulated to minimize total tasks execution delay. To tackle this intractable mixed integer non-convex problem, firstly, we decouple it into separated multiple tasks segmentation and subtasks parallel scheduling problems. Secondly, the multiple tasks segmentation problem is further decomposed into single task segmentation problem, where the optimal task segmentation ratio function is proposed and proved. Thirdly, the Advantage Actor Critic (A2C) algorithm is applied to choose computation node for subtasks parallelly in an online manner for the time-varying network. Finally, the multiple tasks segmentation scheme is incorporated into A2C algorithm to achieve end-to-end joint optimization of tasks segmentation and parallel scheduling with cooperative computing. Simulation results represent the superiority and effectiveness of the proposed algorithm compared with the benchmarks, such as binary tasks offloading.