Energy-efficient mobile edge computing : computation offloading and resource management

Abstract

Driven by the visions of Internet of Things and 5G communications, recent years have seen a paradigm shift in mobile computing, from the centralized Mobile Cloud Computing towards Mobile Edge Computing (MEC). The main feature of MEC is to push mobile computing, network control and storage to the network edges (e.g., base stations and access points) so as to enable computation-intensive and latency-critical applications at the resource-limited mobile devices. By offloading intensive computation from mobiles to proximate edge servers, MEC promises dramatic reduction in computation latency and mobile energy consumption. Achieving these promised gains requires seamless integration of the advanced techniques from two disciplines of wireless communications and mobile computing. This dissertation contributes to the emerging area by investigating the energy-efficient computation offloading and resource management for different MEC systems.

First, for prolonging battery lives of small devices, this dissertation presents a novel design of wirelessly powered single-user MEC system by integrating two promising technologies of computation offloading and microwave power transfer. Assuming binary offloading, we optimize the control policies for both the operation modes of local computing and offloading, and use the results for making the optimal offloading decision. The effects of base station transmission power and the computation deadline on the optimal offloading decision are also characterized and demonstrated by simulation.

Next, to scavenge time-varying idling computational resources at peer mobiles, we design the energy-efficient opportunistic computation offloading from a user to a helper with a dynamically loaded CPU. By exploiting non-causal helper CPU-state information, the policies of adaptive data transmission and energy-efficient data partitioning are developed by using constrained optimization theory. As the estimation for helper CPU-state information may be inaccurate in practice, we further propose robust opportunistic offloading algorithms based on the distribution of dynamic helper CPU. By modeling the channel and helper-CPU as Markov chains, stochastic optimization theory is applied to design both the optimal and sub-optimal offloading controllers.

Last, we investigate the energy-efficient resource management for both the synchronous and asynchronous multiuser computation offloading in MEC systems. Specifically, for synchronous offloading where mobiles share identical data-arrival time instants and computation deadlines, the allocation of radio-and-computational resource is optimized for minimizing the weighted sum mobile energy consumption. The optimal policy based on time-division multiple access is proved to have a threshold-based structure with respect to a derived offloading priority function of mobile state (e.g., local-computing power and channel gain). The result is further extended to the MEC system based on orthogonal frequency-division multiple access. For the asynchronous MEC systems with mobiles having heterogeneous data-arrival time instants and computation deadlines, an efficient iterative algorithm is proposed to compute the optimal offloaded data sizes and offloading durations for the mobiles. We also characterize the effects of the orders of data-arrival time instants and computation deadlines on the optimal scheduling order.