1 A Socially-Aware Hybrid Computation Offloading Framework for Multi-access Edge Computing Shuai Yu, Boutheina Dab, Zeinab Movahedi, Rami Langar, Li Wang Abstract—Computation offloading manages resource-intensive and mobile collaborative applications (MCA) on mobile devices where much processing is replicated with multiple users in the same environment. In this article, we propose a novel hybrid multicast-based task execution framework for multi-access edge computing (MEC), where a crowd of mobile devices at the network edge leverage network-assisted device-to-device (D2D) collaboration for wireless distributed computing (MDC) and outcome sharing. The framework is socially aware in order to build effective D2D links. A key objective of this framework is to achieve an energy-efficient task assignment policy for mobile users. Specifically, we first introduce the socially aware hybrid computation offloading (SAHCO) system model, which combines of MEC offloading and D2D offloading in detail. Then, we formulate the energy-efficient task assignment problem by taking into account the necessary constraints. We next propose a Monte Carlo Tree Search based algorithm, named, TA-MCTS for the task assignment problem. Simulation results show that compared to four alternative benchmark solutions in literature, our proposal can reduce energy consumption up to 45.37%. Index Terms—mobile collaborative application, wireless distributed computing, computation offloading, multicast communication, socially aware, monte carlo tree search. ✦ 1 I NTRODUCTION N OWADAYS, the advances in hardware technology has leaded to more powerful mobile devices in terms of memory, processing speed and network connectivity. This development has pushed to the pervasive computing era, in which different mobile devices, ranging from smart- phones, tablets and laptops to low-power sensors have widely penetrated to our everyday life. Accompanied by the emergence of near-to-eye display technologies, a variety of mobile collaborative applications (MCA) are developed to meet the user’s requirements, such as augmented reality (AR) [2], collaborative gaming [3] and mobile crowd sensing applications [4]. These applications make use of complex al- gorithms for camera tracking, image processing and pattern recognition which are resource-intensive and hence, beyond the capabilities of current mobile devices. To cope with such challenges, a recent promising alternative consists in offloading the whole or some parts of these applications out of the mobile device, specially on powerful cloud servers. Another alternative is to take advantage of available • Shuai Yu is with the School of Data and Computer Science, Sun Yat- sen University, Guangzhou 510275, China, and he was with Laboratoire d’Informatique de Paris 6 (LIP6), Sorbonne Université, Paris 75005, France, (Email:yushuai@mail.sysu.edu.cn). • Boutheina Dab is with the LIP6, Sorbonne Université, Paris 75005, France, (E-mail: Boutheina.Dab@lip6.fr). • Zeinab Movahedi is with the Computer Engineering School of Iran University of Science and Technology (IUST), Tehran, Iran (zmova- hedi@iust.ac.ir). • Rami Langar is with LIGM CNRS-UMR 8049, University Paris Est Marne-la-Vallée (UPEM), Champs-sur-Marne 77420, France, (E-mail: Rami.Langar@u-pem.fr). • Li Wang is with School of Electronic Engineering, Beijing Univ. of Posts and Telecommunications, 100876, China, she is also with Key Labora- tory of the Universal Wireless Communications, Ministry of Education, P.R.China. (E-mail: liwang@bupt.edu.cn). • A preliminary version of this paper appeared in the proceedings of the 2016 Global Communication Conference (GLOBECOM 2016) [1]. nearby mobile resources for executing resource-intensive applications of mobile devices, referred as device-to-device (D2D) computation offloading, Transient Clouds [5], or wireless distributed computing (WDC) [6], [7]. The main asset of computation offloading within a cluster of mo- bile devices consists in reducing the per-node and net- work power, energy, and processing resource requirements. Power analysis has shown that this approach is beneficial over local processing under certain conditions, particularly when the computational cost exceeds the communication overhead [7]. The latter conditions are usually achieved when resource intensive tasks are distributed in short-range networks [6]. Considering the short-range network services, multicast communication plays an important role in data distribution and energy reduction. As a combination of D2D and multicast, computation offloading through D2D multicast communication [8] enables multiple proximate users to share the content items of their common interests and computation results. This latter combination can greatly reduce power consumption (both communication and com- putation) and improve spectral efficiency in a local network. Privacy issue is an important threat for the hybrid com- putation offloading system. On the one hand, mobile users risk exposing their sensitive data (e.g., personal images, personal clinical data and business financial records in real- life situations) by offloading it to untrusted mobile users through D2D computation offloading. On the other hand, MEC computation offloading will also face privacy leakage issues, when mobile users migrate computation to the edge servers for edge data analytic. Thus, it is a huge challenge to protect mobile users’ sensitive information in the hy- brid computation offloading system. A potential solution to address the challenges is to consider the social domain factors besides physical domain in the hybrid computation offloading. It is essential to find a model to describe the