Client devices operating at the edges on the Internet, in homes, cars, offices, and elsewhere, are highly heterogeneous in terms of their hardware configurations, form factors, and capabilities, ranging from small sensors to wearable and mobile devices, to the stationary ones like smart TVs and desktop machines. With recent and future advances in wireless networking allowing all such devices to interact with each other and with the cloud, it becomes possible to combine and augment capabilities of individual devices via services running at the edge - in edge clouds - and/or via services running in remote datacenters. The virtual platform approach to combining and enhancing such devices developed in this research makes possible the creation of innovative end user services, using low-latency communications with nearby devices to create for each end user exactly the platform needed for current tasks, guided by permissions and policies controlled by remote, cloud-resident social network services (SNS). To end users, virtual platforms operate beyond the limitations of individual devices, as natural extensions of those devices that offer improved functionality and performance, with ease-of-use provided by cloud-level global context and knowledge.

Edge-cloud brings the computation power as close to the clients as possible. This reduces latencies and overall computation time in the cloud. Thanks to the mobile nature of clients we must be able to migrate tasks among different servers. The goal of this thesis is to examine possible problems in communication and propose the architecture of framework. Our framework uses gRPC and is written as module to it. It is platform independent, uses reliable communication and focuses on easy usage. We provide implementation of this framework with some example uses. 1

Guaranteeing Quality of Service (QoS) in an (edge-)cloud environment is one of the biggest open problems in the field of cloud computing. Currently, deploy- ment of cloud applications is managed by cloud orchestration systems, such as Kubernetes. These systems make deployment of applications in cloud easier than ever, offering their users the benefits of availability, scalability and resilience. However, at the moment they are not capable of optimizing the deployment of cloud applications with respect to performance QoS metrics, such as response time and throughput. The thesis proposes an approach that provides probabilistic guarantees on the performance QoS metrics in an (edge-)cloud environment. The approach is based on assessing the performance of cloud applications and subsequently controlling their deployment in a way that the applications are deployed only in the environ- ments in which their performance does not violate their QoS requirements. The thesis also presents a proof-of-concept implementation of that approach. The im- plementation verifies the effectiveness of the approach and will serve for further research.

The dynamic nature of the edge cloud and future network infrastructures is another challenge to be added when modeling end-to-end service performance using machine learning. That is, a model that has been trained for one specific environment may see reductions in prediction accuracy over time due to e.g., routing, migration, or scaling decisions. Transfer learning has been proposed as an approach for leveraging already learned knowledge in a new environment, especially when the amount of training data is limited in that new domain. This thesis proposes and evaluates a heterogeneous transfer learning approach via feed-forward neural networks that addresses model transfer across different domains with diverse input features, a natural consequence of network, and cloud infrastructure re-orchestration. Transfer gain is measured, and a positive gain is empirically shown in the vast majority of cases. The impact neural network architectures have on transfer gain is also analyzed, returning interesting insights for several different neural network architectures. The method we propose is evaluated on data traces collected from a testbed that runs a video-on-demand service and a key-value store under two different load conditions. Finally, the social, economic, and environmental impact of the work is discussed, as well as possible future lines of work and the accomplished objectives. / Edge-molnets dynamiska karaktär och framtida nätverksinfrastrukturer är utmaningar som måste tas i beaktande när man modellerar prestanda med hjälp av maskininlärning. Det vill säga, en modell som har tränats för en specifik miljö kan se försämrad noggrannhet över tid på grund av t.ex. routing, migration eller skalningsbeslut i infrastrukturen. Transfer Learning har föreslagits som ett sätt att utnyttja redan inlärd kunskap i en ny miljö, särskilt när mängden träningsdata är begränsad i den nya domänen. Denna uppsats föreslår och utvärderar en heterogen metod för transfer learning som utnyttjar neurala nätverk vilka möjliggör modellöverföring mellan olika domäner med olika features. Överföringsvinsten mäts och en positiv vinst visas i de allra flesta scenarier med empirisk data. De effekter som neurala nätverksarkitekturer har på överföringsvinsten analyseras också, vilket ger intressanta insikter för valet av olika neurala nätverksarkitekturer. Metoden vi föreslår utvärderas på data som samlats in från en testbädd som kör en video-on-demand-tjänst och en nyckelvärdeslagring under två olika lastscenarier. Slutligen diskuteras arbetets sociala, ekonomiska och miljöpåverkan, liksom möjliga framtida arbetslinjer och de uppnådda målen.

The goal of this thesis is to propose a layer on top of edge-cloud, in order to provide soft real-time guarantees on the execution time of applications. This is done in order to satisfy the soft-real time requirements set by the developers of latency-sensitive applications. The proposed layer uses a predictor of execution time, in order to find combinations of processes with, which satisfy the soft real- time requirements when collocated. To implement the predictor, we are provided with information about the resource usage of processes and execution times of collocated combinations. We utilize similarity between the processes, cluster analysis, and regression analysis to form four prediction methods. We also provide a boundary system of resource usage used to filter out combinations exceeding the capacity of a computer. Because the metrics indicating the resource usage of a process can vary in their usefulness, we also added a system of weights, which estimates the importance of each metric. We experimentally analyze the accuracy of each prediction method, the influence of the boundary detection system, and the effects of weights. 1

Modern CPS and mobile applications like augmented reality or coordinated driving, etc. are envisioned to combine edge-cloud processing with real-time requirements. The real-time requirements however create a brand new challenge for cloud processing which has traditionally been best-effort. A key to guaranteeing real-time requirements is the understanding of how services sharing resources in the cloud interact on the performance level. The objective of the thesis is to design a mechanism which helps to categorize cloud applications based on the type of their workload. This should result in specification of a model defining a set of applications which can be deployed on a single node, while guaranteeing a certain quality of the service. It should be also able to find the optimal node where the application could be deployed.

Mobile phones have evolved from feature phones to smart phones with processing power that can compete with personal computers ten years ago. Nevertheless, the computing power of personal computers has also multiplied in the past decade. Consequently, the gap between mobile platforms and personal computers and servers still exists. Mobile Cloud Computing (MCC) has emerged as a paradigm that leverages this difference in processing power. It achieve this goal by augmenting smart phones with resources from the cloud, including processing power and storage capacity. Recently, Mobile Edge Computing (MEC) has brought the benefits from MCC one hop away from the end user. Furthermore, it also provides additional advantages, e.g., access to network context information, reduced latency, and location awareness. This thesis explores the advantages provided by MEC in practice by augmenting an existing application called Human-Centric Positioning System (HoPS). HoPS is a system that relies on context information and information extracted from a photograph of signposts to estimate a user's location. This thesis presents the challenges of enabling HoPS in practice, and implement strategies that make use of the advantages provided by MEC to tackle the challenges. Afterwards, it presents an evaluation of the resulting system, and discusses the implications of the results. To summarise, we make three primary contributions in this thesis: (1) we find out that it is possible to augment HoPS and improve its response time by a factor of four by offloading the code processing; (2) we can improve the overall accuracy of HoPS by leveraging additional processing power at the MEC; (3) we observe that improved network conditions can lead to reduced response time, nevertheless, the difference becomes insignificant compared with the heavy processing required. / Utvecklingen av mobiltelefoner har skett på en rusande takt. Dagens smartphones har mer processorkraft än vad stationära datorer hade för tio år sen. Samtidigt så har även datorernas processorer blivit mycket starkare. Därmed så finns det fortfarande klyftor mellan mobil plattform och datorer och servrar. Mobile Cloud Computing (MCC) används idag som en hävstång för de olika plattformernas processorkraft. Den uppnår detta genom att förbättra smartphonens processorkraft och datorminne med hjälp från datormolnet. På sistånde så har Mobile Edge Computing (MEC) gjort så att förmånerna med MCC är ett steg ifrån slutanvändaren. Dessutom så finns det andra fördelar med MEC, till exempel tillgång till nätverkssammanhangsinformation, reducerad latens, och platsmedvetenhet. Denna tes utforskar de praktiska fördelarna med MEC genom att använda tillämpningsprogrammet Human-Centric Positioning System (HoPS). HoPS är ett system som försöker att hitta platsen där användaren befinner sig på genom att använda sammanhängande information samt information från bilder med vägvisare. Tesen presenterar även de hinder som kan uppstå när HoPS implementeras i verkligheten, och använder förmåner från MEC för att hitta lösningar till eventuella hinder. Sedan så utvärderar och diskuterar tesen det resulterande systemet. För att sammanfatta så består tesen av tre huvuddelar: (1) vi tar reda på att det är möjligt att förbättra HoPS och minska svarstiden med en fjärdedel genom att avlasta kodsprocessen; (2) vi tar reda på att man kan generellt förbättra HoPS noggrannhet genom att använda den utökade processorkraften från MEC; (3) vi ser att förbättrade nätverksförutsättningar kan leda till minskad svarstid, dock så är skillnaden försumbar jämfört med hur mycket bearbetning av information som krävs.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第24261号 / 情博第805号 / 新制||情||136(附属図書館) / 京都大学大学院情報学研究科社会情報学専攻 / (主査)教授 伊藤 孝行, 教授 吉川 正俊, 教授 神田 崇行, 特定准教授 LIN Donghui / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

Edge cloud is a distributed computing architecture that is growing in popularity. It aims to bring the cloud closer to the edge of a network, reducing latency and improving performance through the use of distributed servers (edge nodes) spread out geographically. However, in the case of sudden increases in user requests, a node may run short of resources and need to implement a strategy that allows the node's service to degrade its service quality to a level that requires fewer resources so that the service can still be delivered. One such strategy is brownout, a control theory-based algorithm that dynamically controls the node's service quality in order to meet e.g., a latency goal. This thesis explores the use of brownout, previously used in combination with load balancing in the cloud, in conjunction with load balancing in an edge-cloud environment. In this thesis, four load-balancing strategies are evaluated in a Kubernetes-based edge-cloud environment, along with an application that implements the brownout feature. Two of the strategies are originally designed to be used with brownout but made for the regular cloud, one is a recently introduced strategy that performs well in the edge cloud but is brownout unaware, and the last is a random load balancer used as a baseline (also brownout unaware). The goal of the evaluation is to determine the efficiency of these strategies in different edge-cloud scenarios, with regard to service quality-weighted throughput, average latency, adherence to a set latency goal, and outsourcing (requests load balanced to another edge node). The results show that the first two strategies perform worse than the random load balancer in many regards. Their performance is also less predictable and tends to get worse with increasing network delays. The edge cloud strategy, however, shows an improvement in performance when the brownout is introduced in the majority of the test scenarios.  Furthermore, the thesis introduced three possible modifications to make one of the cloud-based strategies perform better in the edge cloud. These modifications were tested in the same environment as the other load-balancing strategies and compared against each other. The first modification consisted of making the load-balancing logic treat its own node differently from other edge nodes.  The second version was devised to only outsource when a certain resource threshold is exceeded and the last version was designed to prioritize its own node when below a certain resource threshold. The last version improved on the others and performed better than the base version in all measured metrics. Compared to the edge cloud strategy with brownout, it performed better with regard to service quality-weighted throughput but was outperformed in all other metrics.

Nubo is a broker-less and decentralized edge cloud marketplace that provides APIexposure for integration between entities of the marketplace, i.e., service providers and customers of edge cloud. The core functionalities of this marketplace are implemented by using Hyperledger Fabric blockchain. Some of the marketplace features are Subscription to a service, Service registration, Service onboarding, and Application onboarding. Although it has many features, the Nubo marketplace does not provide performance assurance to the customers, which is one of the main reasons for the lack of trust between the service providers and customers. The SLA management can resolve this issue, and it builds a trusted environment in the marketplace. SLA management facilitates the generation of a formal contract between the service provider and the customer, in which the parties agree on the service’s anticipated performance level (measured in terms of QoS indicators). It also involves consequences in the form of a penalty when the expectations are not achieved during the contract period. One of the challenges in SLA management in the marketplace is to describe the SLA considering different edge cloud characteristics such as multi-parties, network services, etc. In this thesis, we studied the specific requirements for describing the edge cloud SLA and SLA intent, and then we proposed TOSCA models that can be used to describe SLA and SLA intent in the marketplace. In the end, we evaluated the proposed models with the help of an edge cloud use case. / <p>Due to the covid pandemic, I made my presentation online.</p>