IEICE TRANS. ??, VOL.Exx–??, NO.xx XXXX 200x 1 INVITED PAPER Mobile Edge Computing Empowers Internet of Things Nirwan ANSARIya) and Xiang SUNyb), SUMMARY In this paper, we propose a Mobile Edge Internet of Things lored for IoT devices is critical to empowering the current (MEIoT) architecture by leveraging the fiber-wireless access technology, IoT system. In addition, only providing interconnections to the cloudlet concept, and the software defined networking framework. The share raw data among IoT devices is not enough to gain the MEIoT architecture brings computing and storage resources close to Internet of Things (IoT) devices in order to speed up IoT data sharing and analytics. insight behind the big IoT data; the insight is more valuable Specifically, the IoT devices (belonging to the same user) are associated to for the society as a whole. Thus, it is beneficial to provision a specific proxy Virtual Machine (VM) in the nearby cloudlet. The proxy the IoT system with a comprehensive cognitive capability VM stores and analyzes the IoT data (generated by its IoT devices) in real- such that high-level knowledge can be extracted from the time. Moreover, we introduce the semantic and social IoT technology in the context of MEIoT to solve the interoperability and inefficient access big IoT raw data streams by applying various types of data control problem in the IoT system. In addition, we propose two dynamic mining and machine learning methods [3]. The data center proxy VM migration methods to minimize the end-to-end delay between infrastructure has been demonstrated to provision resources proxy VMs and their IoT devices and to minimize the total on-grid energy flexibly and efficiently [4]; meanwhile, various parallel com- consumption of the cloudlets, respectively. Performance of the proposed puting architectures and distributed storage frameworks have methods is validated via extensive simulations. key words: Internet of Things, mobile edge computing, cloudlet, semantics, been designed based on a data center (e.g., MapReduce [5] social network, green energy. and Spark [6]). Thus, it is desirable to transmit the big IoT data from the IoT devices to remote data centers via the In- 1. Introduction ternet for further data analysis. Yet, this would place a heavy burden on the network to conduct data aggregation from IoT Internet of Things (IoTs) is enabling interconnections among devices to a centralized data center, and thus exponentially a tremendous number of things such that different things can increase the network delay for transmitting big IoT data to share their observations of the physical world. According remote data centers, especially during the peak time. Note to a new Gartner forecast, 26 billion things (excluding PCs, that delay is a key performance metric in provisioning the tablets, and smartphones) will be installed in 2020 [1]. Cisco Quality of Service (QoS) for many IoT applications. For predicted that 50 billion devices will be connected to the In- instance, smart grid applications have stringent requirement ternet by 2020 [2]. These connected things will generate on latency up to 20 ms; processing automation applications a humongous volume of data, which digitally represent the (i.e., monitoring and diagnosing of industrial elements and states of the physical world. However, owing to the resource processes) imposes latency requirements ranging from 50 constrained nature, many IoT devices cannot always guar- ms to 100 ms [7]. antee the interconnections, i.e., some IoT devices cannot be The Mobile Edge Computing (MEC) concept [8] is es- reachable owing to their periodical sleep schemes and inter- sentially bringing the computing and storage capability from mittent wireless connections. Thus, it is important to design remote data centers to the mobile edge in order to reduce an efficient mechanism to facilitate resource constrained IoT the network delay between end devices and computing re- devices in sharing their data over the network. Also, the re- sources. Specifically, various computing resources, attached source constrained IoT devices cannot feasibly conduct com- to edge routers, wireless access points (WAPs), and smart plicated data access management, thus sharing IoT data over gateways, are available for nearby mobile devices; thus, these the network poses serious security challenges, i.e., unautho- devices can offload their workloads to computing resources rized users/devices may easily access and misuse the shared at the edge, thus potentially reducing the energy consumption IoT data, which may contain personal information. There- of the devices and accelerating the computing processes [9]. fore, designing an efficient access control mechanism tai- Empowering IoT with MEC can essentially improve the QoS for IoT applications. Basically, IoT applications, which try Manuscript received May 1, 2017. to obtain the corresponding data from different types of IoT Manuscript revised June 20, 2017. devices and generate high-level knowledge by analyzing the yThe authors are with Advanced Networking Lab., Helen and acquired data based on data analytic models, would be de- John C. Hartmann Department of Electrical & Computer Engi- ployed at the mobile edge, and thus the data streams gen- neering, New Jersey Institute of Technology, Newark, NJ 07102, erated by the IoT devices would be uploaded to the IoT ap- USA. a) E-mail: [email protected] plications without traversing the mobile core network. This b) E-mail: [email protected] (Corresponding author) can significantly alleviate the traffic load in the core network DOI: 10.1587/trans.E0.??.1 and potentially speed up the IoT applications in processing Copyright © 200x The Institute of Electronics, Information and Communication Engineers IEICE TRANS. ??, VOL.Exx–??, NO.xx XXXX 200x 2 big IoT data streams. In this paper, we will design a novel Mobile Edge 2.2.1 Wired backhaul solutions IoT (MEIoT) architecture by leveraging the Fiber-Wireless (FiWi) access technology, the mobile network, the cloudlet Wired backhaul provides the advantages of high reliabil- concept, and the Software Defined Networking (SDN) ity, high data rate, and high availability. A WAP (such as framework to efficiently share and analyze the big IoT data MBS-1 in Fig. 1) can communicate with the mobile core at the mobile edge. The rest of the paper is organized as network based on a wired connection (e.g., xDSL), which follows. In Sec. II, we describe the proposed MEIoT ar- traverses an access switch. The access switch is connected chitecture. In Sec. III, we illustrate the four challenges of to a cloudlet and conducts L2/L3 switching among the WAP, the current IoT system and propose four potential solutions the cloudlet, and the mobile core network. A cloudlet [11], tailored for the MEIoT architecture. In Sec. IV, we evaluate which comprises a number of interconnected Physical Ma- the performance of the proposed solutions via simulations. chines (PMs), provides computing and storage resources to We briefly delineate the future work in Sec. V, and present IoT devices with low latency. the conclusion in Sec. VI. Passive optical networks (PONs) can potentially provi- sion cloud computing [12], and thus a WAP can also utilize 2. MEIoT architecture the optical backhaul to achieve extra low communications de- lay. For example, MBS-2 in Fig. 1 is connected to an Optical In order to facilitate IoT data sharing and analytics, we pro- Network Unit (ONU), which is further connected to an Op- pose the MEIoT architecture, as shown in Fig. 1. The MEIoT tical Line Terminal (OLT) via an optical splitter/combiner. architecture comprises five parts, i.e., multi-interface wire- The function of an ONU is to aggregate the traffic from its less access network, heterogeneous backhauling, distributed connected WAPs and communicate with its connected OLT cloudlets, hierarchical structure of a cloudlet, and the SDN based on the assigned wavelength channels. The function based mobile core network. We will next detail these five of an OLT is to provide L2/L3 switching between mobile parts. core network and its connected ONUs. Note that there are two different types of ONUs in the MEIoT architecture, tra- ditional ONU and ONU-Cloudlet (ONU-C). Different from 2.1 Multi-interface wireless access network traditional ONUs, an ONU-C, which is normally connected to a local cloudlet, can not only relay the traffic between Various WAPs, such as WiFi access points, static Small Cells its connected OLT and WAPs but also provide the switch- (SC) (e.g., pico cell, femto cell, etc.), dynamic SCs (e.g., ing function to enable the local communications between its drone mounted SCs), and Macro BSs (MBSs), have already WAPs and its connected cloudlet [13]. Thus, the commu- been deployed in the mobile network and provide high radio nications between the local WAPs and the cloudlet can be coverage and network capacity. Thus, distributed WAPshave offloaded from the OLT and the mobile core network. This the potential to connect all IoT devices whether they are mov- can significantly reduce the traffic load of the mobile core ing or static. Yet, different IoT devices have different com- network and the OLT. munications requirements; that is, some energy-sensitive IoT devices (e.g., smart meters) require very low transmission 2.2.2 Wireless backhaul solutions date rate and some energy-insensitive devices (e.g., surveil- lance devices and mobile phones) need high-speed transmis- Wireless backhual technologies present the advantages of sion to meet their embedded application requirements. The flexible deployment and low cost. Currently, many SCs (es- heterogeneous data transmission requirements among IoT pecially for mobile SCs, such as drone mounted SCs [14]) devices effectuate different devices to adopt different wire- apply the wireless backhaul solutions to facilitate the com- less technologies (e.g., D2D communications, NarrowBand munications between MBSs and SCs. The wireless backhaul IoT communications, LTE communications, etc.) to share solutions can be divided into two categories: in-band and their sensed data.