(POSTER)DNN Task Allocation for Edge-Aided IoT

Task allocation and scheduling methods are essential for edge-aided IoT to efficiently execute emerging computing-intensive deep neural network (DNN) applications. However, existing studies mainly overlook the waiting time of task execution on edge server caused by the limited parallel computing capacity. This work proposes a task allocation and scheduling method termed DNN-TA by considering the edge limitation. It first formulates the problem of task allocation and offloading for minimizing the application execution delay as a non-linear programming problem. It then converts the non-linear problem into a mixed integer linear programming problem with higher dimensions to reduce the complexity. Extensive experimental results demonstrate that the proposed DNN-TA method significantly reduces the average application execution delay.