Edge computing and CAV (Connected Autonomous Vehicle) fields can work as a team. With the short latency and high responsiveness of edge computing, it is a better fit than cloud computing in the CAV field. Moreover, containerized applications are getting rid of the annoying procedures for setting the required environment. So that deployment of applications on new machines is much more user-friendly than before. Therefore, this paper proposes a framework developed for the CAV edge computing scenario. This framework consists of various programs written in different languages. The framework uses Docker technology to containerize these applications so that the deployment could be simple and easy. This framework consists of two parts. One is for the vehicle on-board unit, which exposes data to the closest edge device and receives the output generated by the edge device. Another is for the edge device, which is responsible for collecting and processing big load of data and broadcasting output to vehicles. So the vehicle does not need to perform the heavyweight tasks that could drain up the limited power.

Mobile edge computing (MEC) has recently been introduced as a key technology, emerging in response to the increased focus on the emergence of new heterogeneous computing applications, resource-constrained mobile devices, and the long delay of traditional cloud data centers. Although many researchers have studied how the heterogeneous latency-critical application requirements can interact with the MEC system, very few have addressed how to deploy a flexible and scalable MEC infrastructure at the mobile operator for the expected heterogeneous mobile traffic. The proposed system model in this research project relies on the Network Function Virtualization (NFV) concept to virtualize the MEC infrastructure and provide scalable and flexible infrastructure regardless of the underlying physical hardware. In NFV-enabled networks, the received mobile workload is often deployed as Service Function Chains (SFCs), responsible for accomplishing users' service requests by steering traffic through different VNF types and virtual links. Thus, efficient VNF placement and orchestration mechanisms are required to address the challenges of the heterogenous users' requests, various Quality of Service (QoS) requirements, and network traffic dynamicity. This research project addresses the scalable design and dimensioning of an agile NFV-enabled MEC infrastructure problem from a dual perspective. First, a neural network model (i.e., a subset of machine learning) helps proactively auto-scale the various virtual service instances by predicting the number of SFCs required for a time-varying mobile traffic load. Second, the Mixed-Integer Linear Program (MILP) is used to create a physical MEC system infrastructure by mapping the predicted virtual SFC networks to the MEC nodes while minimizing deployment costs. Numerical results show that the machine learning (ML) model achieves a high prediction accuracy of 95.6%, which demonstrates the added value of using the ML technique at the edge network in reducing deployment costs while ensuring delay requirements for different latency-critical applications with high acceptance rates. Due to the exponential nature of this MILP formulation, we also propose a scalable bender decomposition approach with near-optimal results at a significantly reduced design and dimensioning cost. Numerical results show the viability of the bender decomposition approach in its proximity to the optimal dimensioning cost and in its reasonable solution time.

The convergence of mobile computing and cloud computing enables new mobile applications that are both resource-intensive and interactive. For these applications, end-to-end network bandwidth and latency matter greatly when cloud resources are used to augment the computational power and battery life of a mobile device. This dissertation designs and implements a new architectural element called a cloudlet, that arises from the convergence of mobile computing and cloud computing. Cloudlets represent the middle tier of a 3-tier hierarchy, mobile device — cloudlet—cloud, to achieve the right balance between cloud consolidation and network responsiveness. We first present quantitative evidence that shows cloud location can affect the performance of mobile applications and cloud consolidation. We then describe an architectural solution using cloudlets that are a seamless extension of todays cloud computing infrastructure. Finally, we define minimal functionalities that cloudlets must offer above/beyond standard cloud computing, and address corresponding technical challenges.

Vehicular edge computing (VEC) brings the cloud paradigm to the edge of the network, allowing nodes such as Roadside Units (RSUs) and On-Board Units (OBUs) in vehicles to perform services with location awareness and low delay requirements. Furthermore, it alleviates the bandwidth congestion caused by the large amount of data requests in the network. One of the major components of VEC, computation offloading, has gained increasing attention with the emergence of mobile and vehicular applications with high-computing and low-latency demands, such as Intelligent Transportation Systems and IoT-based applications. However, existing challenges need to be addressed for vehicles' resources to be used in an efficient manner. The primary challenge consists of the mobility of the vehicles, followed by intermittent or lack of connectivity. Therefore, the MPR (Mobility Prediction Retrieval) data retrieval protocol proposed in this work allows VEC to efficiently retrieve the output processed data of the offloaded application by using both vehicles and road side units as communication nodes. The developed protocol uses geo-location information of the network infrastructure and the users to accomplish an efficient data retrieval in a Vehicular Edge Computing environment. Moreover, the proposed MPR Protocol relies on both Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication to achieve a reliable retrieval of data, giving it a higher retrieval rate than methods that use V2I or V2V only. Finally, the experiments performed show the proposed protocol to achieve a more reliable data retrieval with lower communication delay when compared to related techniques.

IoT has become tremendously popular. It provides information access, processing and connectivity for a huge number of devices or sensors. IoT systems, however, often do not process the information locally, rather send the information to remote locations in the Cloud. As a result, it adds huge amount of data traffic to the network and additional delay to data processing. The later feature might have significant impact on applications that require fast response times, such as sophisticated artificial intelligence (AI) applications including Augmented reality, face recognition, and object detection. Consequently, edge computing paradigm that enables computation of data near the source has gained a significant importance in achieving a fast response time in the recent years. IoT devices can be employed to provide computational resources at the edge of the network near the sensors and actuators. The aim of this thesis work is to design and implement a kind of edge computing concept that brings AI models to a small embedded IoT device by the use of virtualization concepts. The use of virtualization technology enables the easy packing and shipping of applications to different hardware platforms. Additionally, this enable the mobility of AI models between edge devices and the Cloud. We will implement an AI model inside a Docker container, which will be deployed on a FireflyRK3399 single-board computer (SBC). Furthermore, we will conduct CPU and memory performance evaluations of Docker on Firefly-RK3399. The methodology adopted to reach to our goal is experimental research. First, different literatures have been studied to demonstrate by implementation the feasibility of our concept. Then we setup an experiment that covers measurement of performance metrics by applying synthetic load in multiple scenarios. Results are validated by repeating the experiment and statistical analysis. Results of this study shows that, an AI model can successfully be deployed and executed inside a Docker container on Arm-Cortex-based single-board computer. A Docker image of OpenFace face recognition model is built for ARM architecture of the Firefly SBC. On the other hand, the performance evaluation reveals that the performance overhead of Docker in terms of CPU and Memory is negligible. The research work comprises the mechanisms how AI application can be containerized in ARM architecture. We conclude that the methods can be applied to containerize software application in ARM based IoT devices. Furthermore, the insignificant overhead brought by Docker facilitates for deployment of applications inside a container with less performance overhead. The functionality of IoT device i.e. Firefly-RK3399 is exploited in this thesis. It is shown that the device is capable and powerful and gives an insight for further studies.

Computational offloading advances the deployment of Mobile Edge Computing (MEC) in the next generation communication networks. However, the distributed nature of the mobile users and the complex applications make it challenging to schedule the tasks reasonably among multiple devices. Therefore, by leveraging the idea of Software-Defined Networking (SDN) and Service Composition (SC), we propose a Software-Defined Service Composition model (SDSC). In this model, the SDSC controller is deployed at the edge of the network and composes service in a centralized manner to reduce the latency of the task execution and the traffic on the access links by satisfying the user-specific requirement. We formulate the low latency service composition as a Constraint Satisfaction Problem (CSP) to make it a user-centric approach. With the advent of the SDN, the global view and the control of the entire network are made available to the network controller which is further leveraged by our SDSC approach. Furthermore, the service discovery and the offloading of tasks are designed for MEC environment so that the users can have a complex and robust system. Moreover, this approach performs the task execution in a distributed manner. We also define the QoS model which provides the composition rule that forms the best possible service composition at the time of need. Moreover, we have extended our SDSC model to involve the constant mobility of the mobile devices. To solve the mobility issue, we propose a mobility model and a mobility-aware QoS approach enabled in the SDSC model. The experimental simulation results demonstrate that our approach can obtain better performance than the energy saving greedy algorithm and the random offloading approach in a mobile environment.

Applications for Internet-enabled devices use machine learning to process captured data to make intelligent decisions or provide information to users. Typically, the computation to process the data is executed in cloud-based backends. The devices are used for sensing data, offloading it to the cloud, receiving responses and acting upon them. However, this approach leads to high end-to-end latency due to communication over the Internet. This dissertation proposes reducing this response time by minimizing offloading, and pushing computation close to the source of the data, i.e. to edge servers and devices themselves. To adapt to the resource constrained environment at the edge, it presents an approach that leverages spatiotemporal locality to push subparts of the model to the edge. This approach is embodied in a distributed caching framework, Cachier. Cachier is built upon a novel caching model for recognition, and is distributed across edge servers and devices. The analytical caching model for recognition provides a formulation for expected latency for recognition requests in Cachier. The formulation incorporates the effects of compute time and accuracy. It also incorporates network conditions, thus providing a method to compute expected response times under various conditions. This is utilized as a cost function by Cachier, at edge servers and devices. By analyzing requests at the edge server, Cachier caches relevant parts of the trained model at edge servers, which is used to respond to requests, minimizing the number of requests that go to the cloud. Then, Cachier uses context-aware prediction to prefetch parts of the trained model onto devices. The requests can then be processed on the devices, thus minimizing the number of offloaded requests. Finally, Cachier enables cooperation between nearby devices to allow exchanging prefetched data, reducing the dependence on remote servers even further. The efficacy of Cachier is evaluated by using it with an art recognition application. The application is driven using real world traces gathered at museums. By conducting a large-scale study with different control variables, we show that Cachier can lower latency, increase scalability and decrease infrastructure resource usage, while maintaining high accuracy.

To support increasingly sophisticated sensors and resource-hungry applications with the current-used Lithium-based batteries and to augment mobile computing power further, the concept of the Cloudlet-based offloading is proposed which enables to migrate part of application computing tasks from battery-limited low-capacity mobile elements to the local edge. Such Cloudlet-based offloading technologies extend the provisioning of computing and storage capabilities from remote Cloud Data Centers to the proximity of end users via heterogeneous networks. However, Cloudlet-based offloading is required to coordinate among User Equipment, inter-Cloudlet nodes and remote Cloud Data Centers, which emerges new challenges and issues regarding how to enable Cloudlet-based offloading in the context of mobile edge environment and how to achieve execution- and energy-efficient offloading allocation under limited available resources. In this dissertation, a Cloudlet-based Mobile Cloud offloading prototype is first proposed. A mechanism for handling diverse computing resources is described; by adopting it, idle public resources can be easily configured as additional computing capabilities in the virtual resource pool. A fast deployment model is built to relieve the migration and installation cost when adapting the platform. An energy-saving strategy is utilized to reduce the consumption of computing resources. Security components are implemented to protect sensitive information and block malicious attacks in the cloud. Concerning the limited processing capability on the edge, a task-centric energy-aware Cloudlet-based Mobile Cloud model is formulated. A Cloudlet task-based offloading mechanism is proposed to achieve energy-aware offloading resource preparation and scheduling on the Cloudlet. A Cloud task-centric scheduling algorithm is presented for the green collaborative offloading processing between Cloudlet and remote Cloud. Considering the dynamic and heterogeneity of the offloading environment, a hybrid offloading model to solve the heterogeneous resource-constraint offloading issues on the dynamic Cloudlets. A queue-based offloading framework is developed to formulate and analyze the mixed migration-based and partition-based offloading behaviours on the Cloudlet. The execution and energy-aware heterogeneous offloading resource allocation problem is formalized and solved. A time series-based load prediction model is designed on the Cloudlet to achieve fine-grain proactive resource allocation. Regarding the mobility of User Equipment and the diverse priority of offloading tasks, an edge-based mobility-aware offloading model is modeled to solve the intra-Cloudlet offloading scheduling issue and inter-Cloudlet load-aware heterogeneous resource allocation issue. A priority-based queueing model is designed to formulate the intra-Cloudlet mobility-aware offloading scheduling problem, resolved by a heuristic solution. The energy-aware inter-Cloudlet resource selection procedure is formalized in a mobility-aware multi-site resource allocation model, which is further solved by lightweight dynamic load balancing.

This dissertation first conducts a study on raw-data level cooperative perception for enhancing the detection ability of self-driving systems for connected autonomous vehicles (CAVs). A LiDAR (Light Detection and Ranging sensor) point cloud-based 3D object detection method is deployed to enhance detection performance by expanding the effective sensing area, capturing critical information in multiple scenarios and improving detection accuracy. In addition, a point cloud feature based cooperative perception framework is proposed on edge computing system for CAVs. This dissertation also uses the features' intrinsically small size to achieve real-time edge computing, without running the risk of congesting the network. In order to distinguish small sized objects such as pedestrian and cyclist in 3D data, an end-to-end multi-sensor fusion model is developed to implement 3D object detection from multi-sensor data. Experiments show that by solving multiple perception on camera and LiDAR jointly, the detection model can leverage the advantages from high resolution image and physical world LiDAR mapping data, which leads the KITTI benchmark on 3D object detection. At last, an application of cooperative perception is deployed on edge to heal the live map for autonomous vehicles. Through 3D reconstruction and multi-sensor fusion detection, experiments on real-world dataset demonstrate that a high definition (HD) map on edge can afford well sensed local data for navigation to CAVs.

<div><div><div><p>The increasing cost of cloud services and the need for decentralization of servers has led to a rise of interest in edge computing. In recent years, edge computing has become a popular choice for latency-sensitive applications like facial recognition and augmented reality because it is closer to the end users compared to the cloud. However, the presence of multiple edge servers adversely affects the reliability due to difficulty in maintenance of heterogeneous servers. In this thesis, we first evaluate the performance of various server configuration models in edge computing using EdgeCloudSim, a popular simulator for edge computing. The performance is evaluated in terms of service time and percentage of failed tasks for an Augmented Reality application. We evaluated the performance of the following edge computing models, Exclusive: Mobile only, Edge only, Cloud only; and Hybrid: Edge & Cloud hybrid with load-balancing on the Edge, and Mobile & Edge hybrid. We analyzed the impact of variation of different parameters such as WAN bandwidth, cost of cloud resources, heterogeneity of edge servers, etc., on the performance of the edge computing mod- els. We show that due to variation in the above parameters, the exclusive models are not sufficient for computational requirements and there is a need for hybrid edge computing models. </p><p>Next, we introduce a novel edge computing model called unmanaged edge computing and propose an orchestration scheme in this scenario. Although infrastructure providers are working toward creating managed edge networks, personal devices such as laptops, desktops, and tablets, which are widely available and are underutilized, can also be used as potential edge devices. We call such devices Unmanaged Edge Devices (UEDs). Scheduling application tasks on such an unmanaged edge system is not straightforward because of three fundamental reasons—heterogeneity in the computational capacity of the UEDs, uncertainty in the availability of the UEDs (due to the devices leaving the system), and interference among multiple tasks sharing a UED. In this work, we present I-BOT, an interference-based orchestration scheme for latency sensitive tasks on an Unmanaged Edge Platform (UEP). It minimizes the completion time of applications and is bandwidth efficient. I-BOT brings forth three innovations. First, it profiles and predicts the interference patterns of the tasks to make scheduling decisions. Second, it uses a feedback mechanism to adjust for changes in the computational capacity of the UEDs and a prediction mechanism to handle their sporadic exits, both of which are fundamental characteristics of a UEP. Third, it accounts for input dependence of tasks in its scheduling decision (such as, two tasks requiring the same input data). To demonstrate the effectiveness of I-BOT, we run real-world unit experiments on UEDs to collect data to drive our simulations. We then run end-to-end simulations with applications representing autonomous driv- ing, composed of multiple tasks. We compare to two basic baselines (random and round-robin) and two state-of-the-arts, Lavea [SEC-2017] and Petrel [MSN-2018] for scheduling these applications on varying-sized UEPs. Compared to these baselines, I-BOT significantly reduces the average service time of application tasks. This reduction is more pronounced in dynamic heterogeneous environments, which would be the case in a UEP.</p></div></div></div>