CASOPIS REPUBLICKE AGENCIJE ZA TELEKOMUNIKACIJE

1. INTRODUCTION

The advances in information, computer technology and broadband connectivity have eased the way to reach desired content. TV service is not immune to these changes and watching television is today increasingly switching to an on-demand type of experience. This is especially true for the Internet TV (IPTV) services, where there is a phenomenal growth of demand for interactive TV experience, placing a high burden on the underlying solution that requires numerous precisely orchestrated technologies to deliver a quality viewing experience. It is capable of providing Video-on-Demand (VoD), live TV, private video recording (PVR) capabilities, time-shifted television (TSTV), Near VoD, Subscription VoD services, etc. This transition to the time shifted viewing patterns has also brought forth many technical challenges regarding the distribution of electronic content, such as how to send large video files, how to deal with the stream load when all users are ordering the same or different films, how to transmit the video stream to a global network over a long distance. It is where a Content Delivery Network (CDN) solution comes into the overall VoD system design in order to resolve these issues.

CDN is a distributed video management and delivery system integrated within the data network. It is based on a system of servers with massive storage that places copies from the content library closer to the end user so as to maximize the available bandwidth, and consequently reduces the data access times. Topology of CDN follows central-edge architecture, thus it can push the popular video content to the edge nodes, while utilizing the global load balancing and application redirection technologies to ensure the end-user can watch the requested program without delay.

In this article, we are to shed additional light on the technical challenges surrounding distribution of electronic content, namely designing Video-on-Demand Systems, scaling Content Delivery Networks, and allocating databases in distributed computing systems.

2. STATE OF THE ART

Use of CDNs in context of the Internet is a well known phenomenon; it reduces the content acquisition time by strategically placing servers at the network edge [1]. The content selected according to carefully designed policies, usually based on content popularity in terms of geographies and viewing times is replicated and propagated to the edge servers. Relative to serving all the back end content, this hierarchical decentralized approach brings significant advantages in terms of access time and scalability performance. These benefits will normally come at the cost of using more servers and needing more complex management algorithms to support the content replication, updates and distribution.

There are many challenges in respect of a VoD system design, some touch the system architecture [3], [6], [8], other concern media server design [2], or video distribution [4], [5]. As mentioned earlier, in the context of CDNs, the placement of video servers is an important consideration. Alternatives include centralized video servers [7], centralized video servers with distributed video buffers [9], two- or multitier hierarchical distributed video servers [10], and fully replicated distributed video servers.

P2P topologies might offer scalability and cost effectiveness for certain types of CDN applications. Chord [11-13] uses hash functions to generate identities, but organizes the servers into a ring ordered according to those identifiers, without taking into account the real servers’ locations. Each server has a routing table with the information about the next server and so, when it receives a request, it checks to see whether it has the requested content to fulfill the request. Otherwise, it forwards it through this next server, until the request reaches a destination server that has the content and can fulfill the request.

3. VOD SOLUTION

3.1. Service Architecture

VoD solution integrates the enablers from the entertainment, telecommunication and computer industries and provides video services using clusters of video servers on a broadband network. Users are no longer restricted to being passive watchers. They are allowed to choose the program contents, to decide the viewing schedule, and to interact with the programs with such operations as pause, jump forward, speed-up, etc. Thus, the video cluster must scale up, if needed, to accommodate numerous concurrent user requests to watch and to interact with different parts of the same video, or to allow each user to watch individual videos.

The simplified generic architecture of a VoD system is shown in Figure 1. The business ecosystem consists of the four main players, namely, the network provider, the program provider, the service provider, and the user. The user generates requests to the service provider, who will obtain the necessary material from the program providers and deliver it to the user over the infrastructure provided by the network provider. Thus, the service provider acts as an agent able to access various types of content and aggregate it according to the user’s preferences. It is possible for the network, program, and service providers to be the same entity, but, in general, they are more likely to be distinct organizations. Lately, this business system has been undergoing further evolution, since anyone with marketable content can offer his services directly to the user as Over the Top (OtT) services using the open nature of the Internet.

Figure 1. Simplified Generic VoD Solution Architecture

3.2. Technology and Business Trends

In order to fully understand the dynamics of techno economical forces shaping current solutions for Content Delivery Networks, Video-on-Demand Systems and allocating databases in distributed computing systems, it is very important to take notice of the following trends:

• There is a clear migration towards general purpose computing hardware by VoD vendors. Focus and innovation are in terms of software architecture and service enablers, and progressive vendors are relying on commercial off-the-shelf (COTS) hardware and open API software to maintain their competitiveness.
• The second major shift is the transition from proprietary content management to a stand-alone content delivery network (CDN) addition to the integral VoD content propagation solution. CDN-based content management systems enable rapid ingestion and content propagation to any device from any location. At the same time it optimizes bandwidth utilization by locating and using the most adequate source and deploying transcoding, if needed, to match the capabilities of the device being served.
• The last, but not the least, there is strong evidence of growing importance of the VoD back office in supporting switched digital video (SDV) and on-demand content delivered to wireless devices, in addition to VoD session management,. This expanded role of the back office is setting a trajectory by which the back office systems could become the most valuable component of any VoD system.

New methods of VoD delivery are emerging to combine access to deep content libraries with the appeal of an extensive online marketplace and appear only to be limited by the connection bandwidth of the living room. All VoD solutions are not necessarily on-demand; rather, they are “near” on-demand solutions, depending on the speed of their subscriber’s Internet connection. Increased competition for VoD is driving new functional requirements and a new generation of VoD systems is featuring the following functionalities:

• Access to deeper content libraries – expose all titles within the VoD library in order to enable users gain access to all available titles, be able to easily find and purchase/play them.
• Handle content propagation delays throughout the network – this should shorten currently needed one to three hours to ingest and 11 to 15 hours to propagate content before it is made available to the users.
• Integrate VoD into the Electronic Program Guide (EPG) in order to make navigation through available video programming and selection of VOD titles a unified experience.
• Encrypt VoD streams on per session-basis.
• Manage availability of VoD titles in such way as to give preference to a specific type of content if certain conditions in terms of the resource availability are met.

The most critical parameter for the overall performance of a given VoD solution remains to be the download/display delays. This is to some extent relieved by the continued fierce competition among service providers offering high-speed Internet: larger bandwidth reduces the delivery delay.

IPTV and traditional VoD services are being blended together to combine their respective benefits. Infrastructure components that exclusively serve VoD are being swapped for common ICT platforms and related services can be offered at relatively small incremental cost over the current IPTV solutions.

4. CONTENT DELIVERY NETWORKS

CDN nodes are normally deployed in multiple locations, linked into a virtual network by a web of different backbone connections. The control mechanisms provide logics for cooperation between servers to respond to the requests for content by end users, transparently moving content between the nodes to optimize the delivery process. Optimization can yield reduced bandwidth costs, shorter access/load times or improved global availability of content. The number of CDN nodes and servers varies, depending on the chosen topology; certain deployments number thousands of nodes with tens of thousands of servers in many remote Points of Presence (PoP), while others can build a global network with fewer geographical PoPs.

The Edge Network is grown outward by further acquiring co-location facilities, bandwidth and servers.

Most of current CDNs are built and operated on the Application Service Provider (ASP) business model. Nevertheless, an increasing number of Internet network operators/owners are building their own CDNs to improve on-net content delivery and to generate revenues from content customers.

4.1. Architecture

As shown in Figure 1, a video server system cloud is crucial for the overall design. It is the configuration and topology of this cloud of video servers that form a CDN network that is in turn a critical component for IPTV/VoD solution. Network topology can come in many different flavors that can be classified into three basic categories: centralized, hierarchical decentralized and hybrid. In the centralized topology all requests will be sent to be processed at the same location, usually called Headend. The distributed server system distributes requests to many sites, located closer to the users, thus alleviating the congestion in the network and the bottleneck due to the central server.

Let us now briefly present the most common of CDN topologies based on the location where the VoD servers are placed. Traditional telecommunication networks, as depicted in Figure 2, are based on a hierarchical approach, consisting of three main components: core network, service routing level 1 (SR1) and service routing level 2 (SR2). Users access the services using different access technologies that are not the subject of study here.

Figure 2. Hierarchical Layered structure of a Traditional Communication Network

Figure 3. Hybrid CDN Architecture

Hybrid architecture: Figure 3. depicts a CDN with hybrid architecture. In this topology, video servers are still placed in SR1 and SR2 regions, as shown in Figure 2, butthe servers of the SR2 region are organized, for instance, as a Chord ring, forming P2P groups. Spatial distribution of users is not uniform; for that reason, servers in the SR2 might have different capacities, depending on the number of customers they are responsible for. Sharing the identifier space according to these capacities enables proper resource assignment.

Each SR2 server stores only a partial set of videos and, whenever a viewer requests a video from the assigned SR2 server, the server searches for the video on its peer group, instead of propagating the request up to any of the SR1 servers. This procedure is common across all SR2 servers and consequently reduces the load for SR1 servers. This CDN topology can be further subdivided according to the storage capacity management.

• Single video copy -There is only one copy of the video in the P2P group located at the responsible server. Should a request be issued by another server, the responsible server sends the video directly to the costumer.
• Multiple video copies - Multiple copies of the same video can be stored on the SR2 servers. A single server is still responsible for each video; however, a requested video is sent to another serving server rather than to the consumer directly.

4.2. Performance Metrics

Requests for content are typically algorithmically directed to nodes whose parameters such as location, bandwidth, and storage capacity are optimal for meeting the performance requirements. Performance can be measured in terms of the fewest hops, the shortest time the network takes to respond to the requesting client, or the highest availability in terms of server performance. Cost can be also included into the performance matrix, in which case the algorithm is to prefer PoPs with the lowest cost. Thus, the relevant service performance metrics are:

·VoD Access Latency:It essentially models the VoD session set up time, and represents the elapsed time between the subscriber’s initial request and the start of the VoD on her/his TV. The fraction of VoD requests whose delays exceed a given threshold, derived from the VoD QoS, shall be small and bounded. It is worth noting that operators usually assign higher priority to control packets (i.e., those carrying the network control messages) to reduce the VoD access latency.

·VoD Packet Loss Ratio: It is the measure of the video quality, as well as the key factor affecting a subscriber’s perception of its quality, and is equal to the fraction of video packets lost due to errors or congestion in the network.

·VoD Packet Delay:Along with the packet loss ratio, the VoD packet delay statistics are the key measures of the quality of VoD services, and the user’s perception of them. The packet delay in the network is defined as the traversal time of the packet across the end-to-end infrastructure from the VoD source network layer to the subscriber’s receiver network layer. Since the VoD session supports a real-time service, all packets of each frame shall arrive within a predetermined time in order to be used in reconstructing the video frame at the destination. Those frame packets that arrive too late will be left out and discarded. From the video application point of view, packets that arrive after their deadlines are considered lost in transport, even though the network has actually delivered them to the destination.

·VoD Blocking Probability: It models the service availability, as well as the network ability to deliver it when desired, and is equal to the fraction of VoD requests denied due to the lack of transport resources for delivering the content (more about this later).

Additionally, an operator would like to deliver as many VoD sessions as possible, as long as his network resources (e.g., servers, links, etc.) are efficiently utilized, and the required QoS of VoD services is satisfied. Thus, the relevant network performance metrics are:

·VoD Blocking Probability: As defined before; and

·Resources Utilization: It models the throughput of the network links, and the utilization of the network storage resources. The former represents a fraction of the network link capacity delivering the VoD services, and the latter is the fraction of network storage used/occupied.

Lastly, a VoD operator would like to optimize his storage capacity use. The cache hit rate ratio is the most prevalent metric for measuring and comparing the performance of different caching strategies in the literature. It is defined as follows:

·Cache Hit Ratio: It represents the fraction of total requests that are served from caches.

Since VoD services have stringent QoS requirements, a VoD cache server that intercepts a request for a particular VoD content and has the requested content may decline to serve the request because there is no path between this cache server and the requester with sufficient bandwidth that satisfies all the resource management policies and rules of the network. In other words, all cache hits are not created equal; many cache hits may occur before one of them is able or permitted to deliver the requested content in accordance with the operator’s resource management policies. To reflect this phenomenon, the cache hits are classified into two categories, false cache hits, and positive cache hits. A false cache hit is a hit whose server cannot serve the received VoD request, i.e., this hit is deemed null. A positive cache hit is a hit where there is at least one route between the cache server and the requestor with sufficient capacity to satisfy the QoS of the VoD service, as well as the operator’s resource management rules (e.g., load balancing, etc.). A positive cache hit results in the fulfillment of the request by the cache server. Thus, there is a need to modify the conventional cache hit ratio to define the positive cache hit ratio for VoD services as follows:

·Positive Cache Hit Ratio: It represents the fraction of all VoD requests that are successfully served from VoD cache servers while satisfying the VoD service QoS requirements as well as the resource management rules and policies of the network. Simply put, it is the ratio between the number of positive cache hits and the total number of VoD requests.

Let us further examine the impact of the cache location placement algorithm, and/or the cache replacement rules on the preceding VoD service and network performance metrics, in order to arrive at a reasonable answer. In general, most of the preceding metrics depend on the cache location placement and/or cache replacement algorithms. Since a VoD session will not be set up unless its QoS requirements and the operator’s rules can be satisfied, these algorithms most particularly influence:

–the VoD access latency, the closer to the requestor the better;

–the VoD blocking probability, being further away from the requestor, though having a route with sufficient bandwidth on the way to the requestor may be better than being nearby though facing a congested path to the requestor; and

–the resource utilization, through balancing the load across the network links, servers, etc.

The packet loss ratio and packet delay jitter reflect the inherent characteristics and requirements of the real-time VoD services. They shall not exceed the thresholds required for providing the service with acceptable quality to the viewers. From the operator’s perspective, it is reasonable to say that a VoD network that minimizes the VoD request blocking probability while satisfying the packet loss and maximum access latency constraints is optimum. Recalling the fact that positive cache hit ratio is the fraction of requests successfully served by caches, we reformulate this criterion for defining an “optimal” caching placement and replacement pair of algorithms as follows.

An optimal VoD cache location placement and replacement strategy is the one that maximizes the fraction of VoD sessions served from caches while satisfying the packet loss and maximum access latency constraints.

Let us define the metric, effective cache hit ratio, as the fraction of total requests for VoD sessions that are not blocked (for whatever reason) and are served from caches. Hence, we have

Effective Cache Hit Ratio =Positive Cache Hit Ratio (1.0 – VoD Blocking Probability)

Thus, an optimal caching strategy is the one that maximizes the effective cache hit ratio, while meeting the packet loss and maximum access latency constraints of VoD services. Thinking about it further, effective cache hit ratio is the fraction of the overall VoD network throughput served from the caches. So, we are back to the traditional packet network design optimization problem, i.e., maximizing the throughput of the cache network itself subject to the delay and loss constraints.

5. CONCLUSION

Interactive Television over the Internet (IPTV) is happening at the intersection of television, IT and telephony convergence. Most interactive television systems nowadays are still video-on-demand oriented and the interaction is merely between the audiences and the content servers. Such services are used to deliver programs to the users on request and find many applications in education, entertainment and business. It is in this context that CDNs have emerged as a new technology to overcome the problems arising on the Internet due to the fast growth of the web related traffic, such as slow response times and heavy server loads. CDN is a distributed video management and delivery system integrated within the data network. It is based on a system of servers with massive storage that places copies from the content library closer to the end user so as to maximize the available bandwidth, and consequently reduces the data access times.

The effective cache hit ratio has been defined as a performance measure and combines the throughput of the VoD caches across the network together with the overall utilization of network resources to produce the overall throughput of a given VoD solution. One may define “cache power” as yet another metric for measuring the performance of the caching schemes. The “cache power” metric combines the effective cache hit ratio and the mean access latency of the VoD network into a single metric. It is analogous to the “power” metric in computer networks, and is defined as the ratio between the effective cache hit rate ratio and the mean access latency of a VoD network. Thus, besides the overall utilization of network resources which represents the overall throughput of the network, the effective cache hit ratio should be the prevalent metrics for the CDN performance in context of IPTV/VoD service offerings.

References

1. A. Duta, K. Dutta, H. Thomas, and D. VanderMeer: ”WorldWideWait: a study of Internet scalability and cache-based approaches to alleviate it”, Management Science, Vol. 49, No. 10, October 2003, pp. 1425–1444.
2. Y. N. Dogmata and A. N. Tantawi: “Making a cost-effective video server,” IEEE Multimedia, Winter 1994, pp. 22-30.
3. D. Deloddere, W. Verbiest, and H. Verhille: “Interactive video on demand,” IEEE Commun. Mag, May 1994., pp. 155-162.
4. Papadimitrious et al.: “Multimedia information caching for personalized video-on-demand”, Computer Commun., vol. 18, no. 3, Mar. 1995, pp. 204-216.
5. Paterson et al.: “Introduction to redundant arrays of inexpensive disks (RAID)” , in IEEE COMPCON’89, 1989.
6. Y. H. Chang et al.: “An open-systems approach to video on demand”, IEEE Commun. Mag, May 1994, pp. 68-80.
7. V. P. Rangan and H. M. Vin: “Designing a multiuser HDTV storage server” , IEEE J. Select. Areas Commun., vol. 1, no. 11, 1993.
8. Victor K. Li., Wanjiun Lao, Xiaoxin Qiu, and Eric W. M. Wong: “Performance Model of Interactive Video-on-Demand Systems”, IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, Vol. 14, No. 6, August 1996.
9. Tolga Bektas : “Discrete Location Models for Content Distribution”¸, Centre de recherche sur les transports, Universit´e de Montr´eal, Montr´eal, QC, Canada.
10. G. Peng: CDN: Content Distribution Network, State University of New York at Stony Brook, New York 2006.
11. Ratnasamy S., P. Francis, M. Handley, R. Karp, and S. Shenker: “A scalable content addressable network”, Processings of the ACM SIGCOMM, 2001, pp. 161-172.
12. Pourebrahimi B., K. Bertels, S. Vassiliadis: emphA survey of peer-to peer Networks, Computer Engineering Laboratory, ITS, TU Delft, The Netherlands 2002.
13. I. Stoica., R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan: “Chord: A scalable peer-to-peer lookup protocol for internet applications”, IEEE/ACM Transactions on Networking, vol. 11, no. 1, 2003, pp. 17-32.

Authors

Dragan Bošković received his bachelor’s and master’s degrees from the University of Belgrade in Serbia and his doctorate degree from the University of Bath in the United Kingdom. He is currently Senior Director at the Motorola Inc. Wireless Systems and Networks Research Applied Research Center, where he envisions, develops and drives the convergence of wireless and wireline edge network solutions across Motorola’s businesses, leading a global team of researchers who identify opportunities for new products, systems and services. Prior to this role, Dr. Bošković served as director of Engineering and Technology for Motorola’s Strategic Growth Engine, where he developed technologies that enhanced consumer’s abilities to connect with multimedia and one another. He also worked within Motorola’s R&D sites in Europe for 12 years. Dr. Boscovic has amassed 17 patents, with eight more pending, and has published several papers on topics including wireless technologies and cognitive networks. He is often called upon to share insights on wireless topics at industry events such as the International Consumer Electronic Show, Global Semiconductor Forum, AlwaysOn, Wireless World Initiative and Wireless World Research Forum.

Faramak Vakil received the Ph.D. degree in Electrical Engineering from Columbia University in the City of New York in 1984. He is currently a Distinguished Member of Technical Staff in the Motorola Applied Research Center, working on network and service management. Prior to joining Motorola in 2006, he was a Senior Scientist in the Applied Research Area of Telcordia. Since 1998, Dr. Faramak has almost exclusively been focused on the architecture, design, control and management of mobile Internet. He has devised novel mobility, network self-organization and management, and service creation schemes for mobile Internet. Dr. Faramak is a member of ACM and IEEE. He has published numerous papers and standard contributions, and has several patents.