Article Preview
Top1. Introduction
With the booming of Internet of Things (IoT) and mobile communication technologies, mobile edge (MEC) terminials and nodes are becoming increasingly popular in provisioning computational resources and accessibility to application requesters. MEC supports varying types of computation-intensive applications, such as graphical guided system based on Artificial Intelligence (AI), video gaming, augmented vehicular reality system (Lai, 2018). It (Beck, 2014) helps to transfers the computing task load from the centralized clouds to the edge of the network close to the requesters (Abbas, 2018). In the MEC paradigm, base stations are equipped with a certain amount of computing infrastructures. Thus, they are responsible for both wireless communication and task execution. Mobile requesters are allowed to offloading computational-intensive tasks to the MEC nodes, in this way, requesters can be access these service directly without performing in thire own device or resorting to remote clouds. And that are featured by much less communication overhead and energy consumption than traditional clouds (Xu, 2020 & Chen, 2019).
However, various difficulties, especially the real-time resource allocation, are still to be properly addressed. In a typical edge computing paradigm, MEC servers are usually enhanced with computing components and storage (Wu, 2019). Usually, mobile requesters can move away from of the communication coverage of a certain server which they previsouly contact and lose the communication connection. In case of a connection loss, a requester has to re-contact another MEC node and probably experiences service interruptions which potentially affect to user-perceived quality-of-service (QoS) (Peng, 2019). Therefore, QoS-guaranteed allocation of requesters to suitable MEC nodes with optimized requester coverage rate and minimized reallocation overhead becomes a critical issue.
Most of traditional methods (Lai, 2018 & Peng, 2019 & Yang, 2017 etc.) in this direction are still limited due to the fact they consider offline offloading decision making by employing transient requester positions as model inputs. However, such methods can be ineffective due to the fact that real-world edge requesters are often with high mobility and the offloading actions are thus supposed to be decided in a dynamic way. Instead of considering instantaneous and transient positions of mobile requesters, in this paper, we consider continuing requester trajectories as model inputs and propose a predictive-trajectory-aware online service allocation method. We conduct extensive simulations as well and illustrate that our proposed method outperforms traditional ones in terms of effective service rate and migration overhead.
The remaining part of this paper is organized as follows. Section 2 describes existing methods of task offloading and service allocation in the MEC environment. Section 3 describes and models the service allocation problem . Section 4 presents our method and the simulative results are illustraten in section 5. Finally, in the last section we present some conclusive remarks.