Cost Efficient Offloading Strategy for DNN-Based Applications in Edge-Cloud Environment

Yinhao Huang,Bing Lin,Yongjie Zheng,Junqin Hu,Yuchang Mo,Xing Chen

DOI: https://doi.org/10.1109/ISPA-BDCloud-SustainCom-SocialCom48970.2019.00056

2019-01-01

Abstract:Deep Neural Networks (DNNs) are pervasively used in a large number of applications, such as Microsoft Cortana, Apple Siri, and Google Now. The traditional method is to offload them in the cloud, which causes serious data transmission delay and high network resource cost, as well as user privacy leakage. The emergence of Mobile Edge Computing(MEC) provides a new solution for the execution of DNN-based applications. With the increase of the computing power of mobile devices and the appearance of edge nodes, offloading partial layers of the DNN-based application to the edge will speed up system response, reduce the burden of cloud center and protect privacy of users. Most of existing researches focus on the optimization of response delay for DNN-based applications in edge computing system, but there is no mature solution to the cost optimization of edge computing system. To effectively reduce the cost of edge computing for offloading DNN-based applications, we propose a offloading strategy based on Discrete Particle Swarm Optimization with Genetic Operators(DPSO-GO). By introducing crossover operator and mutation operator of genetic algorithm, it overcomes the defect that the traditional PSO algorithm is easy to fall into local optimum. The results show that our algorithm can reduce the cost of edge computing system effectively while satisfying the delay requirements of all the DNN-based applications.