Jesse E. Simsarian and Marcus Duelk Bell Laboratories, Alcatel-Lucent, Holmdel, NJ, 15th IEEE Workshop on Local and Metropolitan Area Networks, 2007 Chen Bin Kuo ( ) Young J. Won ( )

DPNM Lab.  Introduction  Data and Voice Traffic  IPTV Network Architecture  VoD Content and User Behavior  Caching of Content  Main Traffic  Conclusion 9/11/20152

DPNM Lab.  In the paper, authors analyze the bandwidth requirements in Metropolitan Area Networks (MANs) for providing IPTV services.  Developing a model of IPTV network to determine the optimum location of the cached video content.  Finding out the effect upon MAN traffic by users simultaneously requesting VoD streams 9/11/20153

DPNM Lab.  Applying previous work on content delivery to a realistic metropolitan area work (MAN)  Previous work: unicast content delivery [1][2], proxy server method [3], and multicast groups [4][5]  Considering a service provider is optimizing content delivery  Assuming a small amount of buffering at the client’s set-top box  Delivery stream is viewed nearly in real time  Do not consider VoD multicast [4] 9/11/20154

DPNM Lab.  Developed a cost model of the total network including:  The servers that cache the on-demand video  Costs associated with data switching and transport  Reporting the results about future MAN traffic 9/11/20155

a) Voice Traffic b) Data Traffic 9/11/20156

DPNM Lab.  Voice, video, and data traffic  Different burstiness and flow durations  Different QoS requirement  Network operations and networking equipment will be affected by the future traffic mix 9/11/20157

DPNM Lab.  Flat over the next years  Estimated by market studies forecasting [8]  Fixed landlines (PSTN), VoIP, mobile phone lines 9/11/20158

DPNM Lab.  Download rate during peak hours (6-10 p.m.)  Studies investigating the Internet usage [10][11]  Traffic pattern of more than twenty Internet exchange points (IXPs) worldwide [12] 9/11/20159

DPNM Lab.  Forecasting  Market study from 2005 to 2009 in Western Europe [13]  Predicting annual growth rate of roughly 55% with an increase in the business segment and a growth decline in the consumer market  Summarize [12][14][15][16][17] to derive forecasting numbers  Using lower bound and upper bound to represent high uncertainty  Including P2P traffic to be data traffic 9/11/201510

 IPTV Network Architecture  VoD Content and User Behavior 9/11/201511

DPNM Lab.  Super headend (SHE)  Video hub offices (VHOs) over the wavelength division multiplexed (WDM) national core  Contain VoD servers  Video source offices (VSOs) also contain VoD servers that cache the more popular content  Central offices (COs) 9/11/201512

DPNM Lab. 9/11/201513

DPNM Lab.  Packet routing and switching costs for VoD traffic  C EIF : the cost per bandwidth of Ethernet interfaces  B VHO : the bandwidth of traffic to the VHO server cache  H VHO : the number of packet routing hops this traffic undergoes  B total : the total quantity of VoD traffic  H total : the number of packet hops that all of the VoD traffic undergoes 9/11/201514

DPNM Lab.  The cost of the TDM switching and WDM transport is :  The cost for the VoD servers comes from the cost to store films on disk and the streaming interfaces from the video servers :  C storage is the cost per film for disk storage, and R is the number of films stored at the VSO.  N VSO is the number of VSO nodes, F is the total number of films offered, C stream is the cost per VoD stream, and N sessions is the number of simultaneous VoD sessions 9/11/201515

DPNM Lab.  The authors believe that future VoD offerings will approach the number titles offered  Today’s Netflix DVD mail-delivery service  A catalog of about 60,000 films [19]  Applying Zipf distribution 9/11/201516

 VoD Concurrency [% of Households]  Percent Cached at VSO  Relative Cost & Percent Cached at VSO  Optimum Caching 9/11/201517

DPNM Lab. 9/11/201518

DPNM Lab. 9/11/201519

DPNM Lab.  Using cost model to determine R o, the optimum fraction of content stored at the VSOs. 9/11/201520

DPNM Lab.  Cost curve for  (a) a low concurrency of 1%  (b) a higher concurrency of 10% 9/11/ Intuitively speaking, when the VoD usage becomes high, the content should be delivered closer to the end user

DPNM Lab. 9/11/201522

9/11/201523

DPNM Lab.  (a) the VoD traffic generated in the MAN  (b) VoD portion of total MAN traffic  The voice and data traffic are from the projections of Fig. 1-2 for  Error bars that come form upper and lower bounds of data traffic 9/11/201524

DPNM Lab.  The authors analyzed the bandwidth requirements in the MAN for IPTV systems.  They find that the bandwidth depends on where VoD content caching and video stream delivery is located.  The proposed cost model gives the optimal location for the content, and from that they determine that future networks could have a large percentage of real- time video traffic. 9/11/201525

9/11/201526