A Q-Learning-Driven Data Synthesis and Task Offloading Scheme for Cloud-Edge Collaboration Computing
Y. Liu, Junxiao Ge, Naixue N. Xiong · IEEE Transactions on Consumer Electronics · 2024
Recently, Mobile Edge Computing (MEC) enabled consumer electronics to significantly increase the amount of data available. Such a large volume of consumer-generated data can create new customer value and form the basis for useful and desirable emerging customer applications. It is obvious that large consumer-generated data are stored across geographically distributed edge servers, but most of the previous studies assume the edge server can complete the task independently without obtaining data from other edge servers, which is difficult to apply to the real MEC. Thus, it is an urgent issue for the task offload scheduling scheme to consider that task execution needs to obtain consumer-generated data from other edge servers. In this paper, we attempt to tackle this challenge by proposing a Q-learning-driven Data Synthesis and Task Offloading (Q-DSTO) scheme. First, we establish a problem model where task offloading to edge servers requires obtaining consumer-generated data before execution. Then, we transform the solution of the problem into a Q-learning algorithm framework. Finally, compared with the traditional LRU-TO scheme, experimental results demonstrate that the Q-DSTO scheme reduces the task completion delay by 38.81%, reduces the data transfer volume by 50.51%, and improves network load capacity by 39.31%.