Joint Task Offloading, CNN Layer Scheduling and Resource Allocation in Cooperative Computing System

Abstract

In this paper, we consider a cooperative computing system which consists of a number of mobile edge computing (MEC) servers deployed with convolutional neural network (CNN) model, a remote mobile cloud computing (MCC) server deployed with CNN model and a number of mobile devices (MDs). We assume that each MD has a computation task and is allowed to offload its task to one MEC server where the CNN model with various layers is applied to conduct task execution, and one MEC server can accept multiple tasks of MDs. To enable the cooperative between the MEC servers and the MCC server, we assume that the task of MD which has been processed partially by the CNN model of the MEC server will be sent to CNN model of the MCC server for further processing. We study the joint task offloading, CNN layer scheduling and resource allocation problem. By stressing the importance of task execution latency, the joint optimization problem is formulated as an overall task latency minimization problem. As the original optimization problem is NP hard, which cannot be solved conveniently, we transform it into three subproblems, i.e., CNN layer scheduling subproblem, task offloading subproblem and resource allocation subproblem, and solve the three subproblems by means of extensive search algorithm, reformulation-linearization-technique (RLT) and Lagrangian dual method, respectively. Numerical results demonstrate the effectiveness of the proposed algorithm.