Energy Efficient Computation Offloading in Mobile Edge Computing

Abstract

of the book series: Wireless Networks (WN), 2022: ISBN: 978-3-031-16822-2, 156 pages Reviewer: Bo Rong With the proliferation of mobile devices and the development of Internet of Things (IoT), more and more computation intensive and delay-sensitive applications are running on terminal devices, which results in high-energy consumption and heavy computation load of devices. Due to the size and hardware constraints, the battery lifetime and computing capacity of terminal devices are limited. Consequently, it is hard to process all of the tasks locally while satisfying Quality of Service (QoS) requirements for devices. Mobile Edge Computing (MEC) is considered a promising paradigm that deploys computing resources at the network edge near terminal devices. With the help of MEC, terminal devices can achieve better computing performance and battery lifetime while ensuring QoS. The introduction of MEC also brings the challenges of computation off-loading and resources management under energy-constrained and dynamic channel conditions. It is important to design energy efficient computation offloading strategies while considering the dynamics of task arrival and system environments. This book provides a comprehensive introduction to energy efficient computation offloading and resource management for MEC, covering task offloading, channel allocation, frequency scaling, and resource scheduling. Chapter one introduces the development history, characteristics , and typical applications of mobile cloud computing (MCC) and MEC. The development of mobile computing technology and the popularity of smart mobile devices has led to the widespread deployment and attention of computation-intensive applications. Limited by their own resources, mobile devices can reduce their own burden by offloading some computing tasks. Due to its own characteristics, MCC cannot provide good services for mobile devices, so MEC is introduced to better meet the needs of mobile devices for computation offloading. However, various challenges are encountered in computation offloading, which are summarized and corresponding solutions are given in this book. Chapter two studies the optimization problem of transmission energy consumption in MEC computation offloading. Specifically , with the rapid development of the IoT, more and more computation-intensive applications appear and generate a large number of tasks. However, due to the limitations of the mobile device itself, they cannot fully realize local computation. For this reason, the task can be offloaded to the MEC server at the edge of the network for computing. However, offloading these tasks to edge servers for processing consumes a lot of energy. Therefore, the task offloading optimization problem is formulated with the goal of minimizing transmission energy consumption. However, since the statistics of the mobile device's task generation process and the state of the wireless channel cannot be accurately predicted, stochastic optimization techniques are employed to address this issue. This chapter designs an energy efficient dynamic computation offloading (EEDCO) algorithm based on stochastic optimization techniques that do not require prior statistics on task generation and arrival processes. The EEDCO algorithm can be implemented in a low-complexity parallel manner. And it is verified by experiments that the EEDCO algorithm can minimize the transmission energy consumption while reducing the queue length. Chapter three addresses energy efficient computation off-loading and frequency scaling for IoT devices in MEC. For IoT devices in MEC systems, one of the main challenges is the limited battery capacity to support the processing of computation-intensive tasks for a long time. When a mobile device processes computing tasks locally, the amount of energy consumed is mainly related to the local CPU cycle frequency. When a mobile device performs task offloading, the energy consumed is mainly related to the number of offloaded tasks and the transmission power of the device. In addition, the device will still run for a period of time (tail time) after completing the data transmission, for which tail energy consumption must also be considered. Similarly, the generation and arrival process of wireless channel status and tasks is random and difficult to predict accurately. Therefore, stochastic optimization techniques are employed to transform the original joint task assignment and frequency scaling problem into a deterministic optimization problem. Then, a computation offloading and frequency scaling for energy efficient (COFSEE) algorithm is proposed for IoT devices in MEC systems. Chapter four proposes a delay-aware and energy efficient computation offloading solution based on deep reinforcement learning (DRL) for dynamic MEC systems with multiple edge servers. Mobile devices dynamically generate computing tasks and offload them to edge servers for processing through wireless channels. However, mobile devices tend to choose edge servers that can provide them with the best computing resources. At the same time, due to the delay-sensitive characteristics of mobile devices, in order to complete more tasks and minimize system energy consumption under delay constraints, an end-to-end deep reinforcement learning (E2E_DRL) algorithm is proposed. E2E_DRL algorithm jointly optimizes the allocation of computing resources and the selection of edge servers to maximize long-term cumulative rewards. Experiments show that the E2E_DRL algorithm can realize energy efficient computation offloading and maximize system utility. Chapter five proposes a solution for multi-task energy efficient computation offloading in multi-access MEC via non-orthogonal multiple access (NOMA). This chapter considers two cases. One is that the status of the channel is static. To address this situation, a layered algorithm is proposed to jointly optimize NOMA transmission, resource allocation, and multi-access multi-task computation offloading. The second is that the channel power gain from the mobile device to the edge serv-er is time-varying. To solve the problems caused by different channel implementations in dynamic scenarios, an online algorithm based on DRL is proposed. The algorithm can obtain an approximately optimal offloading scheme through continuous learning, avoiding the joint optimization problem of solving each channel realization. Chapter six concludes the book and gives directions for future research. In conclusion, this book discusses energy efficient computation offloading and resource allocation for MEC deeply. However, the introduction of MEC provokes challenges under energy-constrained and dynamic conditions. Therefore, it is very important to design a strategy for energy efficient computation offloading and resource allocation. To this end, this book discusses issues, such as task offloading, channel allocation, frequency scaling, and resource scheduling in MEC. The presented computation offloading and energy management solutions and the corresponding research results in this book can provide some valuable insights for practical applications of MEC and motivate new ideas for future MEC-enabled IoT networks.