Examination Committee

Prof Danny Hin Kwok TSANG, ECE/HKUST (Chairperson)
Prof Khaled BEN LETAIEF, ECE/HKUST (Thesis Supervisor)
Prof An LIU, ECE/HKUST

 
Abstract

As an emerging paradigm that distributes cloud computing capability to the network edge, mobile edge computing (MEC) supports computation-hungry and latency-sensitive mobile applications and is a key component in 5G mobile networks. Due to the limited resources in MEC systems, it is critical to develop effective joint radio and computational resource management policies. However, most existing works focus on single-objective resource allocation for MEC systems, which aims either to minimize task latency with energy budgets or to minimize device energy consumption with latency constraints, but not both. However, task latency and mobile energy consumption are both critical aspects of users' experience.
 
We first address the multi-objective resource allocation problem for multi-user single-server MEC systems by adopting the system utility, which is a normalized weighted combination of the time and energy saving achieved by computation offloading, as the performance metric. A low-complexity ranking-based algorithm is proposed based on the modified Newton method and the concept of computation offloading priority. Simulation results show that our proposed algorithm achieves a near-optimal performance and greatly outperforms a baseline algorithm with random spectrum allocation. 
 
More general multi-user multi-server MEC systems are considered, where inter-cell interference cannot be neglected and server selection brings a new challenge. To solve this mixed-integer nonlinear programming problem effectively, we propose two algorithms, named the Heu-Con and the Heu-Appro algorithm, respectively, both based on heuristic task offloading algorithms. To solve the resource allocation problem with a given offloading policy, the Heu-Con algorithm uses a conventional method with the help of a convex concave procedure, while the Heu-Appro algorithm considers an approximate problem. It is shown that these two proposed algorithms achieve a performance gain compared with a baseline algorithm that ignores inter-cell interference in its design.