Caching at the network edge has emerged as a viable solution for alleviating the severe capacity crunch in content-centric next generation 5G wireless networks by leveraging localized content storage and delivery. Caching generally works in two phases namely (i) storage phase where parts of popular content is pre-fetched and stored in caches at the network edge during time of low network load and (ii) delivery phase where content is distributed to users at times of high network load by leveraging the locally stored content. Cache-aided networks therefore have the potential to leverage storage at the network edge to increase bandwidth efficiency. In this dissertation we ask the following question - What are the theoretical and practical guarantees offered by cache aided networks for reliable content distribution while minimizing transmission rates and increasing network efficiency? We furnish an answer to this question by identifying fundamental Shannon-type limits for cache aided systems. To this end, we first consider a cache-aided network where the cache storage phase is assisted by a central server and users can demand multiple files at each transmission interval. To service these demands, we consider two delivery models - (i) centralized content delivery where demands are serviced by the central server; and (ii) device-to-device-assisted distributed delivery where demands are satisfied by leveraging the collective content of user caches. For such cache aided networks, we develop a new technique for characterizing information theoretic lower bounds on the fundamental storage-rate trade-off. Furthermore, using the new lower bounds, we establish the optimal storage-rate trade-off to within a constant multiplicative gap and show that, for the case of multiple demands per user, treating each set of demands independently is order-optimal. To address the concerns of privacy in multicast content delivery over such cache-aided networks, we introduce the problem of caching with secure delivery. We propose schemes which achieve information theoretic security in cache-aided networks and show that the achievable rate is within a constant multiplicative factor of the information theoretic optimal secure rate. We then extend our theoretical analysis to the wireless domain by studying a cloud and cache-aided wireless network from a perspective of low-latency content distribution. To this end, we define a new performance metric namely normalized delivery time, or NDT, which captures the worst-case delivery latency. We propose achievable schemes with an aim to minimize the NDT and derive information theoretic lower bounds which show that the proposed schemes achieve optimality to within a constant multiplicative factor of 2 for all values of problem parameters. Finally, we consider the problem of caching and content distribution in a multi-small-cell heterogeneous network from a reinforcement learning perspective for the case when the popularity of content is unknown. We propose a novel topology-aware learning-aided collaborative caching algorithm and show that collaboration among multiple small cells for cache-aided content delivery outperforms local caching in most network topologies of practical interest. The results presented in this dissertation show definitively that cache-aided systems help in appreciable increase of network efficiency and are a viable solution for the ever evolving capacity demands in the wireless communications landscape. / Ph. D. / Caching at the network edge has emerged as a viable solution for alleviating the severe capacity crunch in content-centric next generation 5G wireless networks by leveraging localized content storage and delivery. Caching generally works in two phases namely (<i>i</i>) <i>storage phase</i> where parts of popular content is pre-fetched and stored in caches at the network edge during time of low network load and (<i>ii</i>) <i>delivery phase</i> where content is distributed to users at times of high network load by leveraging the locally stored content. Cache-aided networks therefore have the potential to leverage storage at the network edge to increase bandwidth efficiency. In this dissertation we study cache-aided systems from an information theoretic perspective and identify fundamental Shannontype limits for such systems. The results presented in this dissertation show definitively that cacheaided systems help in appreciable increase of network efficiency and are a viable solution for the ever evolving capacity demands in the wireless communications landscape.

Today’s Internet content providers primarily use Content Delivery Networks (CDNs) to deliver content to end-users with the aim to enhance their Web access experience. Yet the prevalent commercial CDNs, operating in isolation, often face resource over-provisioning, degraded performance, and Service Level Agreement (SLA) violations, thus incurring high operational costs and limiting the scope and scale of their services. / To move beyond these shortcomings, this thesis sets out to establish the basis for developing advanced and efficient content delivery solutions that are scalable, high performance, and cost-effective. It introduces techniques to enable coordination and cooperation between multiple content delivery services, which is termed as “CDN peering”. In this context, this thesis addresses five key issues ― when to peer (triggering circumstances), how to peer (interaction strategies), whom to peer with (resource discovery), how to manage and enforce operational policies (re-quest-redirection and load sharing), and how to demonstrate peering applicability (measurement study and proof-of-concept implementation). / Thesis Contributions: To support the thesis that the resource over-provisioning and degraded performance problems of existing CDNs can be overcome, thus improving Web access experience of Internet end-users, we have: / - identified the key research challenges and core technical issues for CDN peering, along with a systematic understanding of the CDN space by covering relevant applications, features and implementation techniques, captured in a comprehensive taxonomy of CDNs; / - developed a novel architectural framework, which provides the basis for CDN peering, formed by a set of autonomous CDNs that cooperate through an interconnection mechanism, providing the infrastructure and facilities to virtualize the service of multiple providers; / - devised Quality-of-Service (QoS)-oriented analytical performance models to demonstrate the effects of CDN peering and predict end-user perceived performance, thus facilitating to make concrete QoS performance guarantees for a CDN provider; / - developed enabling techniques, i.e. resource discovery, server selection, and request-redirection algorithms, for CDN peering to achieve service responsiveness. These techniques are exercised to alleviate imbalanced load conditions, while minimizing redirection cost; / - introduced a utility model for CDN peering to measure its content-serving ability by capturing the traffic activities in the system and evaluated through extensive discrete-event simulation analysis. The findings of this study provide incentive for the exploitation of critical parameters for a better CDN peering system design; and / - demonstrated a proof-of-concept implementation of the utility model and an empirical measurement study on MetaCDN, which is a global overlay for Cloud-based content delivery. It is aided with a utility-based redirection scheme to improve the traffic activities in the world-wide distributed network of MetaCDN.

The focus of this dissertation is the task of resource allocation in multi- server systems arising from two applications – multi-channel wireless com- munication networks and large-scale content delivery networks. The unifying theme behind all the problems studied in this dissertation is the large-scale nature of the underlying networks, which necessitate the design of algorithms which are simple/greedy and therefore scalable, and yet, have good perfor- mance guarantees. For the multi-channel multi-hop wireless communication networks we consider, the goal is to design scalable routing and scheduling policies which stabilize the system and perform well from a queue-length and end-to-end delay perspective. We first focus on relay assisted downlink networks where it is well understood that the BackPressure algorithm is stabilizing, but, its delay performance can be poor. We propose an alternative algorithm - an iterative MaxWeight algorithm and show that it stabilizes the system and outperforms the BackPressure algorithm. Next, we focus on wireless networks which serve mobile users via a wide-area base-station and multiple densely deployed short- range access nodes (e.g., small cells). We show that traditional algorithms that forward each packet at most once, either to a single access node or a mobile user, do not have good delay performance and propose an algorithm (a distributed scheduler - DIST) and show that it can stabilize the system and performs well from a queue-length/delay perspective. In content delivery networks, each arriving job can only be served by servers storing the requested content piece. Motivated by this, we consider two settings. In the first setting, each job, on arrival, reveals a deadline and a subset of servers that can serve it and the goal is to maximize the fraction of jobs that are served before their deadlines. We propose an online load balanc- ing algorithm which uses correlated randomness and prove its optimality. In the second setting, we study content placement in a content delivery network where a large number of servers, serve a correspondingly large volume of con- tent requests arriving according to an unknown stochastic process. The main takeaway from our results for this setting is that separating the estimation of demands and the subsequent use of the estimations to design optimal content placement policies (learn-and-optimize approach) is suboptimal. In addition, we study two simple adaptive content replication policies and show that they outperform all learning-based static storage policies. / text

Content delivery networks (CDNs) depend on distributed cache servers to reduce the content delivery distance and latency to end users. Nevertheless, a CDN's cache footprint is greatly limited by the high cost in deploying and maintaining large- scale cache servers. To break the limit, CDN providers adopt a new content caching strategy that allows end users to share their storage/bandwidth resources with each other. Two core questions need to answer in this CDN strategy: (1) how to incentivize end users to contribute their resources? and (2) how to facilitate transparent, secure content exchange among end users? We propose a new CDN solution, called SmartSharing, where users contribute their Over-the-top (OTT) devices as mini cache servers. With SmartSharing, an OTT device can share the content the OTT owner is downloading and in addition can cache content for neighboring OTT devices in the same area. To incentivize end users to contribute their resources, SmartSharing uses game theory and the Expectation- Maximization (EM) algorithm to determine content delivery schedule and the pricing scheme. To facilitate content trading among end users, SmartSharing uses smart contracts in Ethereum to create a transparent and safe transaction platform. We evaluate SmartSharing with real-world trace driven simulation as well as smart con- tract prototype in Ethereum using content meta-data and the derived pricing scheme. By disclosing the internal dynamics in content delivery schedule and pricing scheme and analyzing the overhead in content trading, we show that SmartSharing is an effective new CDN solution that benefi ts content providers, CDN, and end users. / Graduate

Internet-scale dissemination of streaming contents (e.g. live sport games) is most successfully being provided by Multicast Content Delivery Networks (Multicast CDN). Multicast CDN is composed of dedicated servers placed strategically over the Internet, which forward content from origin site to end users. Multicast CDN delivers huge amount of data traffic, and therefore its major operational cost is the ISP cost for network access. Existing Multicast CDNs route user requests to most suitable server based on application performance, such as network delay, server throughput, Internet path congestion etc., without taking into account the potentially high ISP cost it may incur. Multicast CDNs need to control their ISP cost to remain commercially competitive since ISP cost is the most indicative factor affecting the pricing of their services. In this work, we present novel Multicast CDN Request Routing algorithms that minimize ISP cost while still maintaining good network performance for users. Multicast CDN Request Routing algorithms control majority of traffic assigned to servers and therefore directly impacts the ISP cost. ISP cost and user network performances are orthogonal metrics of performance and in order to balance the trade-off between them we introduce overall delay as a constraint to the Multicast CDN Request Routing algorithm. Multicast CDNs are business customers of ISPs and therefore can independently choose to reduce their bills by considering the ways in which ISPs charge. We have designed Request Routing algorithms that exploit the economy of scale in ISP charging function in assigning users to servers. We have developed Request Routing algorithms that exploit the nature of Percentile-based charging used by ISPs to compute the charging volume for traffic generated at server sites. Multicast CDN can cooperate with ISPs to reduce the operational cost of both the parties. Multicast CDN controls how traffic is redirected on the overlay, and that can conflict with how underlying ISPs want the traffic to be forwarded. We have developed Request Routing algorithms that assign users to servers that are available over cheaper IP routes. This reduces the transmission costs for ISPs and these savings can be transferred onto Multicast CDN.

L’augmentation constante de la consommation de vidéos par des services par contournement(Over-The-Top) met à mal l’architecture actuelle d’Internet. Alors qu’une écrasante majorité de labande passante aujourd’hui est allouée à la livraison de contenus, les acteurs de l’Internet, tels queles fournisseurs de contenus, les réseaux de distribution de contenus et les fournisseurs d’accès sontobligés d’optimiser leurs réseaux pour supporter la qualité d’expérience attendue par l’utilisateurfinal. Ces réseaux couteux et spécialisés participent à l’ossification de l’Internet, rendant l’évolutionde son architecture plus difficile à moyen terme. En effet, un choix trop spécifique de dimensionnementdes liens et de la localisation des middle-boxes peut être un frein à une mise à jour ultérieureen vue du support de nouveaux cas d’utilisation. Heureusement, les technologies de virtualisationrécemment promues par les grands opérateurs Internet, les vendeurs de solutions et les organismesde standardisation, permettent une réelle programmabilité du réseau et une plus grande versatilitédans les usages de nouveaux équipements. En effet, l’agilité apportée par ces technologies permet ledéploiement de Fonctions Réseaux Virtuelles (VNF, Virtual Network Functions) pouvant s’exécutersur des serveurs de série à bas coût. Quant au Software-Defined Networking , il rend possible unegestion du réseau logiquement centralisée permettant la programmation des commutateurs.L’objectif de cette thèse est de montrer comment la distribution de contenu peut être amélioréecollaborativement à l’aide de la programmabilité de réseaux. Tout d’abord, nous proposons CDNaaS,une solution complète de réseau de livraison de contenu déployée sur une plateforme de "fonctionsréseaux en tant que service" adoptée et évaluée à large échelle dans le cadre du projet Européen FP7T-NOVA. Nous précisons les interfaces, l’architecture et les choix de conception fait pour développerla plateforme afin de fournir performance, auto-dimensionnement et réutilisabilité. Puis, nous proposonsdeux modèles de collaboration permettant aux acteurs de la livraison de contenu de travaillerensemble afin d’augmenter la qualité d’expérience pour l’utilisateur final, tout en promouvant unecompétition saine et une répartition équilibrée de la valeur ajoutée. Finalement, nous étudions lesdéfis liés à l’allocation de ressources virtuelles dans le cas d’un service vCDN, et proposons plusieursheuristiques et algorithmes permettant l’optimisation du coût du service.Cette thèse ouvre la voie à une distribution de contenu collaborative permettant aux utilisateursd’accéder à leurs contenus avec un haut standard de qualité, tout en contribuant à un développementsain de l’Internet. / The constant rise of Over-The-Top video consumption nowadays challenges the current Internetarchitecture. As an overwhelming majority of the bandwidth today is dedicated to the deliveryof video contents, Internet actors such as Content Providers, Content Delivery Network, InternetService Providers are forced to optimize their networks to support the Quality of Experience expectedby the End-Users . Such costly specialized networks participate to the so-called "ossification" of theInternet whichmakes the architecture harder to update, as the placement and dimensioning of linksand middleboxes may be hard to change in the future to support new use cases. Fortunately, thetrend of Virtualizating and Softwarizing the networks pushed by major Telco operators, vendors andstandardization bodies has given hopes that the computing and networking infrastructure can beeasily re-purposed. The agility promoted by technologies such as Network Function Virtualizationand Software Defined Networking makes it possible for middleboxes to be deployed as VirtualNetwork Functions that can run on "commercial off-the-shelf" hardware while having the networkmanaged by a logically centralized controller deploying network configurations on programmableforwarding devices.The goal of this thesis is to show how content distribution can be collaboratively improved thanksto Network Softwarization. First, we propose CDN-as-a-Service (CDNaaS), a complete solution tovirtualize a Content Delivery Network on top of a VNF-as-a-Service platform, adopted and evaluatedat large scale in the FP7 T-Nova European project. We elaborate on the interfaces, architecture anddesign choices made to implement the platform to support performance, automatic scaling andre-usability. Then, we propose two collaboration models allowing the content delivery actors to workjointly on improving End-User Quality of experience while fostering a healthy competition and a fairbalance of revenue. Finally, we study the challenges of NFV resource allocation for the vCDN serviceand propose several heuristics and algorithms to optimize the proposed solution in a cost-effectiveway. This thesis paves the way towards a collaborative content distribution allowing End-Users toaccess their content with the highest standards while contributing to a sound development of theInternet.

Significant portion of today’s internet traffic emerge from multimedia services. When coupled with growth in number of users accessing these services, there is tremendous increase in network traffic. CDNs aid in handling this traffic and offer reliable services by distributing content across different locations. The concept of virtualization transformed traditional data centers into flexible cloud infrastructure. With the advent of cloud computing technology, multimedia providers have scope for establishing CDN using network operator’s cloud environment. However, the main challenge while establishing such CDN is implementing a cost efficient and dynamic mechanism which guarantees good service quality to users. This thesis aims to develop, implement and assess the performance of a model that coordinates deployment of virtual servers in the cloud. A solution which dynamically spawns and releases virtual servers according to variations in user demand has been proposed. Cost-based heuristic algorithm is presented for deciding the placement of virtual servers in OpenStack based federated clouds. Further, the proposed model is implemented on XIFI cloud and its performance is measured. Results of the performance study indicate that virtual CDNs offer reliable and prompt services. With virtual CDNs, multimedia providers can regulate expenses and have greater level of flexibility for customizing the virtual servers deployed at different locations.

The Cloud infrastructure has become an ideal platform for large-scale applications, such as Video-on-Demand (VoD). As VoD systems migrate to the Cloud, new challenges emerge. The complexity of the Cloud system due to virtualization and resource sharing complicates the Quality of Experience (QoE) management. Operational failures in the Cloud can lead to session crashes. In addition to the Cloud, there are many other systems involved in the large-scale video streaming. These systems include the Content Delivery Networks (CDNs), multiple transit networks, access networks, and user devices. Anomalies in any of these systems can affect users’ Quality of Experience (QoE). Identifying the anomalous system that causes QoE degradation is challenging for VoD providers due to their limited visibility over these systems. We propose to apply end user QoE in the management and control of large-scale VoD systems in the Cloud. We present a QoE-based management and control systems and validate them in production Clouds. QMan, a QoE based Management system for VoD in the Cloud, controls the server selection adaptively based on user QoE. QWatch, a scalable monitoring system, detects and locates anomalies based on the end-user QoE. QRank, a scalable anomaly identification system, identifies the anomalous systems causing QoE anomalies. The proposed systems are developed and evaluated in production Clouds (Microsoft Azure, Google Cloud and Amazon Web Service). QMan provides 30% more users with QoE above the “good” Mean Opinion Score (MOS) than existing server selection systems. QMan discovers operational failures by QoE based server monitoring and prevents streaming session crashes. QWatch effectively detects and locates QoE anomalies in our extensive experiments in production Clouds. We find numerous false positives and false negatives when system metric based anomaly detection methods are used. QRank identifies anomalous systems causing 99.98% of all QoE anomalies among transit networks, access networks and user devices. Our extensive experiments in production Clouds show that transit networks are the most common bottleneck causing QoE anomalies. Cloud provider should identify bottleneck transit networks and determine appropriate peering with Internet Service Providers (ISPs) to bypass these bottlenecks.

One of the main challenges in delivering content over the Internet today is the absence of a centralized monitoring and control system [38]. Software Defined Networking has paved the way to provide a much needed control over network traffic. OpenFlow is now being standardized as part of the Open Networking Foundation, and Software Defined Exchanges (SDXes) provide a framework to use OpenFlow for multi-domain routing. Prototype deployments of Software Defined Exchanges have recently come into existence as a platform for Future Internet Architecture to eliminate the need for core routing technology used in today’s Internet. In this work, we look at how application delivery, in particular, Dynamic Adaptive Streaming over HTTP (DASH) and Nowcasting take advantage of a Software Defined Exchange. We compare unsophisticated controllers to more sophisticated ones which we call a ”load balancer” and find that implementing a good controller for inter-domain routing can result in better network utilization and application performance. We then design, develop and evaluate a prototype for a Content Distribution Network (CDN) that uses resources at SDXes to provide higher quality bitrates for a DASH client.

The flexibility of future mobile networks exploiting modern technologies such as cloud-optimized radio access and software-defined networks opens a gateway to deploying dynamic strategies for live and on-demand content delivery. Traditional live broadcasting systems are spectral inefficient. It takes up a lot more radio spectrum than that of mobile networks, to cover the same size of an area. Furthermore, content caching at base stations reduces network traffic in core networks. However, numerous duplicated copies of contents are still transmitted in the unicast fashion in radio access networks. It consumes valuable radio spectrum and unnecessary energy. Finally, due to the present of numerous mobile receivers with a wide diversity of wireless channels in a base station coverage area, it is a challenge to select a proper modulation scheme for video broadcasting to optimize the quality of services for users. In this thesis, the challenges and the problems in the current strategies for content delivery are addressed. A holistic novel solution is proposed that considers user preferences, user mobility, device-to-device communication, physical-layer resource allocation, and video quality prediction. First, a system-level scheduling framework is introduced to increase the spectral efficiency on broadcasting live contents onto mobile networks. It considers the audience preferences for allocating radio resources spatially and temporally. Second, to reduce the redundant transmissions in radio access networks, a content distribution system that exploits user mobility is proposed that utilizes the urban-scale user mobility and broadcasting nature of wireless communication for delay-tolerant large size content. Third, to further reduce the energy consumption in network infrastructure, a content distribution system that relies on both user mobility, and device-to-device communication is proposed. It leverages the mobile users as content carriers to offload the heavy mobile traffic from network-level onto device-level. Fourth, to mitigate the multi-user channel diversity problem, a cross-layer approach is deployed to increase the video quality for users especially for those who have a low signal-to-noise ratio signal. Finally, data mining techniques are employed to predict video qualities of wireless transmissions over mobile networks. The holistic solution has been empirically developed and evaluated. It achieves high spectral and energy efficiency and mitigates the video quality degradation in mobile networks.