Energy-Constrained Partial Offloading in Data Processing Unit (DPU)-Enabled Mobile Edge Computing
Jinyu Liu, Tianhao Lin, Yi Zhang, Yalong Li · 2022
As a dedicated data-centric processor, data processing unit (DPU) can release central processing unit (CPU) from the heavy computation load of data-intensive applications such as data encryption and decryption. In mobile edge computing (MEC), the integration of DPU structure into the MEC server can significantly reduce the processing delay of security-critical tasks. This paper studies the partial offloading problem oriented toward a DPU-enabled MEC system, in which the DPUs are responsible for all computation burden of data encryption/decryption. The objective is to minimize the completion time of security-critical tasks under an energy constraint. We introduce a new EDA-based metaheuristic (NEDA) to solve the resultant optimization problem. By using a task sequence to represent an offloading solution, this algorithm employs a task dispatching strategy to make partial offloading decision for each task in the sequence such that the solution quality can be evaluated. To enhance NEDA’s solution exploration capability, we develop a new population updating mechanism that relies on the current best solution and a probability matrix to generate a new population of high-quality solutions. Experimental results demonstrate NEDA’s effectiveness in solving the partial offloading problem in terms of task completion time.