Virtual Edge Computing Using Vehicular Micro Clouds Falko Dressler * , Gurjashan Singh Pannu * , Florian Hagenauer * , Mario Gerla † , Takamasa Higuchi ‡ and Onur Altintas ‡ * Heinz Nixdorf Institute and Dept. of Computer Science, Paderborn University, Germany † Computer Science Department, University of California - Los Angeles, Los Angeles, CA ‡ Toyota InfoTechnology Center U.S.A., Mountain View, CA {dressler,pannu,hagenauer}@ccs-labs.org, gerla@cs.ucla.edu, {ta-higuchi,onur}@us.toyota-itc.com Abstract—We will discuss the challenges and opportunities of the connected cars vision in relation to the need for distributed data management solutions ranging from the vehicle to the mobile edge and to the data centers. As a novel concept, vehicular micro clouds have been proposed that bridge the gap between fully distributed vehicular networks based on short range vehicle to vehicle communication and cellular based infrastructure for centralized solutions. We will discuss the need for vehicular micro clouds, followed by the architecture, formation of micro clouds, and feasibility of micro clouds. Furthermore, we will cover aspects of efficient data upload and download between cars and a data center facilitated by our micro cloud concept. Index Terms—Vehicular Cloud, Mobile Edge Computing, Vir- tual Cloud Architecture, Vehicular Networking I. I NTRODUCTION Cars today are equipped with a rich set of computing, data storage, communication, and sensor resources in their on- board computer unit. Based on current vehicular networking standards, not only Vehicle to Infrastructure (V2I) but also Vehicle to Vehicle (V2V) communication are supported. Thus, it is believed that cars will play a major role in future Information and Communication Systems (ICT) systems for supporting applications like Intelligent Transportation Systems (ITS) and full-scale smart cities [1]–[3]. Recently, several groups of researchers independently pro- posed the concept of Vehicular Clouds (VCs), which brings the mobile cloud model (or the edge computing model [4]) to vehicular networks. Eltoweissy et al. [2] and Gerla [5] were among the first to propose using a VC as the backbone of ITS, smart cities, and smart electric power grids. Lee et al. [3] drafted important design principles of building VCs on top of Vehicular Ad Hoc Networks (VANETs) in combination with information-centric networking. This concept has been refined by Dressler et al. [1] and Hagenauer et al. [6], [7]. They discussed how to provide cloud- like computation and networking service distributed among cars in both a parking lot as well as using geographic clustering techniques for moving vehicles. Independently, Florin et al. [8] and Arif et al. [9] also explored how to distribute MapReduce- like computation onto cars in a parking lot. Conceptually, these works extend the Mobile Edge Com- puting (MEC) concept [4], [10], which is about bringing computing and storage capabilities from the cloud to the edge of the network, mainly to reduce latencies – MEC is now being standardized as a key component of upcoming 5G networks [11], [12]. Several advantages of VC can be identified: Vehicular Clouds allow diverse resources to be pooled and deployed dynamically to serve users with different needs (e.g., on-demand market for computation and communication service from nearby cars), they further enable autonomy in real-time service sharing and management with lower network latency (if a VC is deployed in close proximity), and VCs are typically decentralized and peer-to-peer, which avoids a single point of failures or market monopoly. Despite all these efforts, there are still many challenges and open problems. Prior work has discussed issues ranging from networking [3] to engineering [13] to security [14]. In this paper, we focus on the issues in the intersection of VC and distributed computing, particularly on issues related to the dynamic and distributed nature of vehicles. To address some of the challenges, we introduce the new paradigm of a virtual cloud architecture. Building upon the Car4ICT concept proposed by Altintas et al. [15], which describes a V2V- based mechanism for providing services from individual cars, we define the concept of a hierarchical Macro-Micro-Cloud (MMC). The main challenges in this context are the management of clusters of cars, i.e., the micro cloud, the data management within the micro cloud and the upload of context information to a backend data center, as well as the download of information from such backend and the dissemination to all interested cars in the geographically local vicinity. We present first solutions to these challenges, which we believe will pave the road for next generation virtual cloud processing using cars as a main ICT resource. The rest of this paper is structured as follows: We first introduce our virtual cloud architecture in Section II. In Section III, we discuss how vehicular micro clouds, i.e., clusters of cars, are maintained both in a stationary case using parked cars and in a very dynamic environment using map-based clustering techniques. We also present concepts for the efficient data download and upload in Sections IV and V , respectively. We finally provide some conclusions in Section VI.