Energy-efficient content delivery networks using cluster shutdown

Content delivery networks (CDNs) deliver Web content to end-users from a large distributed platform of Web servers hosted in data centers belonging to thousands of Internet service providers (ISPs) around the world. The bandwidth cost incurred by a CDN is the sum of the amounts it pays each ISP for routing traffic from its servers located in that ISP out to end-users. A large enterprise may also contract with multiple ISPs to provide redundant Internet access for its origin infrastructure using technologies such as multihoming and mirroring, thereby incurring a significant bandwidth cost across multiple ISPs. This paper initiates the formal study of bandwidth cost minimization in the context of a large enterprise or a CDN, a problem area that is both algorithmically rich and practically very important. First, we model different types of contracts that are used in practice by ISPs to charge for bandwidth usage, including average, maximum, and 95th-percentile contracts. Then, we devise an optimal offline algorithm that routes traffic to achieve the minimum bandwidth cost, when the network contracts charge either on a maximum or on an average basis. Next, we devise a deterministic (resp., randomized) online algorithm that achieves cost that is within a factor of 2 (resp., e/(e - 1)) of the optimal offline cost for maximum and average contracts. In addition, we prove that our online algorithms achieve the best possible competitive ratios in both the deterministic and the randomized cases. An interesting theoretical contribution of this work is that we show intriguing connections between the online bandwidth optimization problem and the seemingly-unrelated but well-studied ski rental problem where similar optimal competitive ratios are known to hold. Finally, we consider extensions for contracts with a committed amount of spend (CIRs) and contracts that charge on a 95th-percentile basis

Internet delivery infrastructures are traditionally optimized for low-latency traffic, such as the Web traffic. However, in recent years we are witnessing a massive growth of throughput-oriented applications, such as video streaming. These applications introduce new tradeoffs and design choices for content delivery networks (CDNs). In this paper, we focus on understanding two key design choices: (1) What is the impact of the number of CDN's peering points and server locations on its aggregate throughput and operating costs? (2) How much can ISP-CDNs benefit from using path selection to maximize its aggregate throughput compared to other CDNs who only have control at the edge? Answering these questions is challenging because content distribution involves a complex ecosystem consisting of many parties (clients, CDNs, ISPs) and depends on various settings which differ across places and over time. We introduce a simple model to illustrate and quantify the essential tradeoffs in CDN designs. Using extensive analysis over a variety of network topologies (with varying numbers of CDN peering points and server locations), operating cost models, and client video streaming traces, we observe that: (1) Doubling the number of peering points roughly doubles the aggregate throughput over a wide range of values and network topologies. In contrast, optimal path selection improves the CDN aggregate throughput by less than 70\%, and in many cases by as little as a few percents. (2) Keeping the number of peering points constant, but reducing the number of location (data centers) at which the CDN is deployed can significantly reduce operating costs.

Content Delivery Networks (CDNs) have emerged as a dominant mechanism to deliver content over the Internet. Despite their importance, to our best knowledge, large-scale analysis of CDN cache performance is lacking in prior literature. A CDN serves many content publishers simultaneously and thus has unique workload characteristics; it typically deals with extremely large content volume and high content diversity from multiple content publishers. CDNs also have unique performance metrics; other than hit ratio, CDNs also need to minimize network and disk load on cache servers. In this paper, we present measurement and analysis of caching workload at a large commercial CDN. Using detailed logs from four geographically distributed CDN cache servers, we analyze over 600 million content requests accounting for more than 1.3 petabytes worth of traffic. We analyze CDN workload from a wide range of perspectives, including request composition, size, popularity, and temporal dynamics. Using real-world logs, we also evaluate cache replacement algorithms, including two enhancements designed based on our CDN workload analysis: N-hit and content-aware caching. The results show that these enhancements achieve substantial performance gains in terms of cache hit ratio, disk load, and origin traffic volume.

Over the last years, the content delivery network (CDN) market has been witnessing globally an increase in the development of cloud-based CDNs, as they constitute a viable and cost-effective alternative to traditional commercial CDNs. Towards that direction, Social-Aware Virtual Network Embedding (SAViNE) for wireless content delivery within the OpenLab project aims at establishing, assessing, and prototyping a novel framework for deploying a CDN over the wireless cloud. Specifically, the goal of SAViNE is to provide experimental validation of the efficiency of social network analysis (SNA) - inspired surrogate server placement strategies, devised for fostering content delivery. The proposed framework has been integrated with the OpenLab experimental facilities, providing a run-time environment for CDN deployment, operation, and performance evaluation over the wireless environment. The framework’s feasibility and scalability is experimentally validated. The study is complemented with repeatability evaluation of experimental measurements, related to the resulting CDN’s operational efficiency.

With the increasing popularity of the Internet video streaming services (e.g. YouTube and Netflix), content delivery networks (CDNs) are heavily used to stream video contents to users, and consume more and more power and bandwidths in recent years. In this paper, we investigate the problem of saving a video streaming CDN’s operating expense, including both its energy cost and the traffic cost. From our measurement study on the CDN infrastructure of Youku, which is the largest video site in China, we find that there exists an inherent conflict between improving a video streaming CDN’s energy efficiency for power saving, and maintaining the CDN’s ISP-friendly server selection policy. To address this problem, we propose a cost-aware capacity provisioning algorithm, which dynamically plans the service capacities of a CDN’s server clusters in numerous ISPs, and optimizes its overall operating cost regarding both the energy consumptions and the cross-ISP traffics. By using the workload derived from real-world measurement and applying actual power and bandwidth price parameters, we show with experiments that our approach can significantly reduce a video streaming CDN’s overall operating cost, and avoid frequent server switches effectively. To our best knowledge, this work is the first one that identifies and resolves the inherent conflict between a CDN’s energy efficiency and its ISP-friendly policy.

The connotation of the cloud resources have been extended to be multi-scale resources, which includes central resources as presented by data center, edge resources as presented by Content Delivery Network (CDN) and end resources as presented by Peer-to-Peer (P2P). Under the development situation of the scale of the cloud services, it is difficult to provide services (e.g. streaming distribution) with guaranteed QoS only relying on single type of resource (e.g. central resources) to geo-distributed users. Therefore, making multi-resources cooperative to provide reliable services is necessary. However, it is a great challenge to realize Federated Management of Multi-scale Resources (FMMR). In this research, we propose the idea of prediction-based FMMR, and present the problem formulation introducing economic profit from the perspective of CDN operators. Then, we present the method of Time-series Prediction based on Wavelet Analysis (TPWA) to predict the resource requirements of streaming cloud services in CDN. Finally, the predictability of the resource requirement pattern of the streaming cloud service and the effectiveness of our proposed method have been verified, based on the traces collected from a real CDN entity.

LofoSwitch: An online policy for concerted server and disk power control in content distribution networks

Planet-scale video Content Delivery Networks (CDNs) deliver a significant fraction of the entire Internet traffic. Effective caching at the edge is vital for the feasibility of these CDNs, which can otherwise incur significant monetary costs and resource overloads in the Internet.We analyze the challenges and requirements for video caching on these CDNs which cannot be addressed by standard solutions. We develop multiple algorithms for caching in these CDNs: (i) An LRU-based baseline solution to address the requirements, (ii) an intelligent ingress-efficient algorithm, (iii) an offline cache aware of future requests (greedy) to estimate the maximum caching efficiency we can expect from any online algorithm, and (iv) an optimal offline cache (for limited scales). We use anonymized actual data from a large-scale, global CDN to evaluate the algorithms and draw conclusions on their suitability for different settings.

It is our great pleasure to welcome you to the Demo Track of WWW 2016, The 25th International World Wide Web Conference, held in Montreal, Canada, during April 11-15, 2016.The WWW 2016 Demo Track, like in the tradition of WWW Demo conference series, allows researchers and practitioners to demonstrate new systems in a dedicated session. Demo contributions are based on an implemented and tested system that pursues one or more innovative ideas in the interest areas of Web data and information management, Web search, Web intelligence tools, Web mining, social network applications and so forth. Topics of interest for the 2016 edition's conference include (but are not limited to) the following ones: Behavioral Analysis and PersonalizationBig Data on the WebCrowdsourcing Systems and Social MediaContent AnalysisGraph Data Management and MiningHigh-Performance Infrastructures for Data- Intensive Web TasksInternet Economics and MonetizationPervasive Web and MobilitySecurity and PrivacySemantic WebSocial Networks and Graph AnalysisWeb Information RetrievalWeb Infrastructure: Datacenters, Content Delivery Networks, and Cloud ComputingWeb MiningWeb ScienceWeb Search Systems and ApplicationsDemo contributions come from academic researchers, industrial practitioners with prototypes or inproduction deployments, as well as from any W3C-related activities. All have in common to show innovative use of Web-based techniques.The WWW 2016 Demo Track call for papers attracted 65 submissions from all over the world (USA, North America, South America, Europe, Australia, Asia, Africa). The program committee reviewed and accepted a very selected collection of 29 papers, and the final statistics is the following: WWW 2016 Demo Track Statistics Number of Submitted Papers 65 Number of Accepted Papers 29.

A common approach when setting up a website is to utilize third party Web hosting and content delivery networks. Without taking this trend into account, any measurement study inspecting the deployment and operation of websites can be heavily skewed. Unfortunately, the research community lacks generalizable tools that can be used to identify how and where a given website is hosted. Instead, a number of ad hoc techniques have emerged, e.g., using Autonomous System databases, domain prefixes for CNAME records. In this work we propose Pythia, a novel lightweight approach for identifying Web content hosted on third-party infrastructures, including both traditional Web hosts and content delivery networks. Our framework identifies the organization to which a given Web page belongs, and it detects which Web servers are self-hosted and which ones leverage third-party services to provide contents. To test our framework we run it on 40,000 URLs and evaluate its accuracy, both by comparing the results with similar services and with a manually validated groundtruth. Our tool achieves an accuracy of 90% and detects that under 11% of popular domains are self-hosted. We publicly release our tool to allow other researchers to reproduce our findings, and to apply it to their own studies.

Cloud computing becomes a popular trend in recent years. In the meanwhile, content delivery network has been used for many years to distribute content all over the world. The relation between cloud computing and content delivery network is very interesting. In this paper, we proposed a cloud based content delivery network. This architecture integrates cloud infrastructure and content delivery network system together. It provides a low cost and elastic web serving cloud platform. Besides, it also provides easy way to manage data for web content developers, and accelerate the speed for users to access web sites. This system uses Hadoop Distributed File System as a cloud infrastructure service, and creates many data clusters around the world. Therefore, content providers could place content to closest node from end users. The result shows a high performance and flexible solution to build a personal content delivery network over cloud.

Content Delivery Networks (CDNs) deliver much of the world's web, video, and application content on the Internet today. A key component of a CDN is the mapping system that uses the DNS protocol to route each client's request to a ``proximal'' server that serves the requested content. While traditional mapping systems identify a client using the IP of its name server, we describe our experience in building and rolling-out a novel system called end-user mapping that identifies the client directly by using a prefix of the client's IP address. Using measurements from Akamai's production network during the roll-out, we show that end-user mapping provides significant performance benefits for clients who use public resolvers, including an eight-fold decrease in mapping distance, a two-fold decrease in RTT and content download time, and a 30% improvement in the time-to-first byte. We also quantify the scaling challenges in implementing end-user mapping such as the 8-fold increase in DNS queries. Finally, we show that a CDN with a larger number of deployment locations is likely to benefit more from end-user mapping than a CDN with a smaller number of deployments.

Content Delivery Networks (CDNs) are a key en-abler for the distribution of large amounts of data with high capacity and low latency. For instance, content streaming companies extensively use geographical distribution and replication to meet the ever-growing demand for media. Optical networks are the only future-proof technology available that meets the reach and capacity requirements of CDNs. However, the robustness of optical networks becomes a point of concern, as they can be a target of deliberate link cuts that can severely degrade network connectivity and cause large-scale service disruption. To mitigate the vulnerabilities, actions can be taken in the optical and/or cloud infrastructures. The replication of content across geographically diverse data centers results in an intrinsic increase of content accessibility. At the network infrastructure level, robustness to attacks can be enhanced by increasing the topology connectivity through link addition. This work focuses on the latter approach and its effectiveness in increasing content accessibility in the presence of deliberate link cuts. The paper proposes a framework for evaluation and enhancement of content accessibility in CDNs by sparse link addition. First, a content accessibility measure called μ-ACA is introduced to gauge the content accessibility of a given network topology under a set of link cut attack scenarios. Based on this measure, a new link addition strategy is defined aimed at maximizing the content accessibility for a given number of extra links. Simulation results on real-world reference topologies show that the proposed strategy can significantly improve content accessibility by adding a very limited number of optical fiber links.

Multicast routing from a set of data centers in elastic optical networks

Enterprise and cloud data centers are comprised of tens of thousands of servers providing petabytes of storage to a large number of users and applications. At such a scale, these storage systems face two key challenges: (a) hot-spots due to the dynamic popularity of stored objects and (b) high reconfiguration costs of data migration due to bandwidth oversubscription in the data center network. Existing storage solutions, however, are unsuitable to address these challenges because of the large number of servers and data objects. This paper describes the design, implementation, and evaluation of Ursa, which scales to a large number of storage nodes and objects and aims to minimize latency and bandwidth costs during system reconfiguration. Toward this goal, Ursa formulates an optimization problem that selects a subset of objects from hot-spot servers and performs topology-aware migration to minimize reconfiguration costs. As exact optimization is computationally expensive, we devise scalable approximation techniques for node selection and efficient divide-and-conquer computation. Our evaluation shows Ursa achieves cost-effective load balancing while scaling to large systems and is time-responsive in computing placement decisions, e.g., about two minutes for 10K nodes and 10M objects.KeywordsLoad balancingstorageoptimizationlinear programming