Multicast Re-examined

Transcription

1 Multicast Re-examined A Contrarian View To Delivering All IP Video A Technical Paper prepared for SCTE/ISBE by Erica Robinson Principal Consultant IBB Consulting Group 1628 JFK Blvd., Suite 170 Philadelphia, PA erobinson@ibbconsulting.com 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved.

2 Title Table of Contents Page Number Introduction 4 Multicast Overview 4 1. Reference mabr Architecture 4 2. Multicast vs. Unicast 6 3. Benefits of Multicast on the Access Network 7 4. Benefits on the CDN 7 Shifting Landscape 8 1. Next Generation Networks Evolving the Network Challenges and Estimated Timeline 9 2. Shifts in Viewership IP Penetration Channel Consolidation Uptake of HEVC Uptake of 4K and HDR Putting it all Together Likely Senarios and Impact to the Network Base Case Short-term Scenario Mid-term Scenario Long-term Scenario Costs to Deploy multicast 18 Conclusion 19 Abbreviations 19 Bibliography & References 20 List of Figures Title Page Number Figure 1 Multicast Reference Architecture 5 Figure 2 Unicast and Multicast Workflows 6 Figure 3 Node Size Projections 9 Figure 4 Live TV as Default TV Source 10 Figure 4 Peak Concurrency Projections 11 Figure 6 IP Penetration Projections 11 Figure 7 Factor Overlay 14 Figure 8 Access Network Bandwidth Required Per Number or Unique Channels Watched Society of Cable Telecommunications Engineers, Inc. All rights reserved. 2

3 List of Tables Title Page Number Table 1 Access Network Downstream and Uptream Demand 6 Table 2 CDN Downstream Traffic 8 Table 3 Scenario Assumptions 15 Table 4 Base Case Scenario 15 Table 5 Short-term Scenario 16 Table 6 Mid-term Scenario 17 Table 7 Long-term Scenario Society of Cable Telecommunications Engineers, Inc. All rights reserved. 3

4 Introduction When multicast was first introduced to ease access network demands during the transition to all IP video, it gained a lot of attention. Widespread cable industry acceptance of using multicast to deliver IP video to the primary screen is based on the assumption that today s linear concurrency rate and node constraints will continue into the future. What this assumption overlooks is that a shift to multicast delivery would occur contemporaneously with the deployment of new techniques in networking, encoding, resolution and historic changes in video consumption habits. Heavy MSO investment in next generation network initiatives such as DOCSIS 3.1 and node splits will decrease node size and increase spectrum per node. At the same time, HEVC, 4K and HDR are starting to gain traction. Subscribers are changing the way they consume video, increasingly opting for more time shifted content versus linear viewing. Additionally, when MSOs deploy IP Set-Top Boxes (STB), a phased approach will be used, similar to other major transitions to new platforms. The combined effects of these developments may change the relative benefit of multicast over unicast, but not universally across cable operators due to different stages of technology uptake. This paper reviews a multicast reference architecture and the industry developments that impact the case for and against multicast. The paper also examines specific examples to offer insight into what MSOs can expect to gain by deploying multicast in the access network. Multicast Overview Multicast is not a new concept for cable operators, which have been using the technology for years to distribute digital video over IP backbone networks to multiple head-ends and hubs to feed broadcast QAMs. There are a range of benefits to consider when using multicast to deliver linear video over the access network to subscriber homes as part of adaptive bit rate (ABR) video delivery strategies. called multicast ABR (mabr), this approach enables IP video clients in the same node to consume a common linear video stream over the access network, thus reducing the access network bandwidth requirements over unicast delivery (where a separate stream is delivered to each subscriber). mabr offers a similar benefit to broadcast QAM used to deliver linear video today over cable HFC networks. Switched Digital Video (SDV) is used as an optimization for broadcast QAM to enable access to more linear channels (long-tail) than can be broadcast given spectrum constraints. Typically with QAM, popular channels are broadcast constantly and long-tail channels are in an SDV pool to only broadcast when at least one subscriber in a node is watching it. mabr can provide a further optimization over SDV since all linear channels can be in the dynamic pool. 1. Reference mabr Architecture There are three incremental key components in a reference mabr architecture: 1. Multicast Server: retrieves ABR content via HTTP requests to the CDN for linear IP video fragments and generates a file multicast for delivery over the access network 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 4

5 2. Multicast Client: as a function of the Home Gateway, joins multicasts over the access network via IGMP and caches IP video fragments in the home for use by ABR players (acting as a CDN edge cache) 3. Multicast Controller: signaling component that receives viewership messages from Multicast Client and instructs the multicast server on whether a file multicast should be generated Figure 1 Multicast Reference Architecture A reference flow is described as follows: 1. IP Player in the home requests a linear channel via standard HTTP requests (e.g., HLS or DASH) 2. Multicast Client works to fulfill the requests with two key actions: a. Forwards requests to the CDN as a proxy enabling fast channel change to receive a manifest and start filling the player buffer via unicast b. Informs the Multicast Controller that an IP Player is tuned to a specific linear channel in a specific format and bit rate 3. Multicast Controller receives the notification and uses business rules to determine if it should generate a file multicast for that channel/format/bit rate (rules varied by operator): a. If/when available, Multicast Client joins the IP file multicast for the requested channel/format/bit rate via IGMP join b. Multicast Client starts side loading fragments in the Home Gateway cache from the multicast instead of forwarding the unicast request to the CDN and informs the Multicast Controller that the stream is being watched (heartbeat) 4. IP Player continues to request fragments and is served from the home gateway cache (looking like unicast in the home to the IP Player) 5. When the IP Player tunes away from the linear channel assuming there are no other IP Players playing the same stream, Multicast Client issues an IGMP leave and informs the Multicast Controller that the IP Player no longer requires the channel 6. Multicast Controller receives the notification and uses business rules to determine if it should stop generating a file multicast for the channel/format/bit rate (rules varied by operator) 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 5

6 2. Multicast vs. Unicast Figure 2 captures the sequence flow for unicast and multicast linear IP video delivery. For both unicast and multicast, initial requests are the same. The IP player sends a manifest request to the home gateway. If the cache within the home gateway does not have the requested manifest, the request must traverse the access network and retrieve the manifest from the CDN. Once the manifest is retrieved, the IP Player will follow the same flow to retrieve the first few fragments. The key difference between unicast and multicast is that unicast has to follow this sequence for every fragment. However, with multicast, once the first few fragments are retrieved, the Multicast Server (based on business rules from the Multicast Controller) will multicast all subsequent fragments. Figure 2 Unicast and Multicast Workflows With unicast, every HTTP request and response will traverse the access network and use bandwidth on the access network. With multicast, the access network bandwidth will equate to only a few HTTP requests, a unique IGMP join per home gateway and one response per channel requested within a node. Therefore, we can summarize the access network downstream and upstream demand in unicast and multicast below: Table 1 Access Network Downstream and Uptream Demand Access Network Downstream Unicast The number of peak streams per node x the average streaming bit rate Multicast The number of unique channels watched per node x the average streaming bit rate 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 6

7 Access Network Upstream The number of peak streams x the number of concurrent HTTP requests x the size of each request The number of peak streams x size of each IGMP join + HTTP manifest request 3. Benefits of Multicast on the Access Network Multicast can, in theory, reduce the bandwidth required on the access network to deliver linear IP video. The reduction comes from subscribers in the same node sharing a video stream (like QAM broadcast) vs. delivering a unique stream (e.g., VOD). To calculate multicast benefits on the access network, we can compare the downstream and upstream demand. We further define number of peak streams per node and number of unique channels watched per node below: Number of peak streams per node. This will be derived from the size of each node and the linear peak concurrency. In the case of local DVR, this will also include the peak concurrency for recordings. Number of unique channels watched per node. This will be derived from the size of each node and the concentration of viewership across channels. Today, in a node size of 250 cable subscribers, we see around 80% linear peak concurrency. Linear peak concurrency is defined as the total number simultaneous linear streams per household at peak. Depending on the MSOs, the IP STB penetration could be anywhere from 0-5%. If IP penetration was instead 100% and all of those 250 cable subscribers are IP video subscribers, we could calculate the benefit of IP multicast over unicast by looking at number of peak streams per node compared with the number of unique channels watched per node. We will refer to this as our base case throughout this paper. In the base case, the most unicast streams would be 200 streams (250 subscribers per node x 80% linear concurrency). If the number of unique channels watched are less than 200, there will be savings in bandwidth on the access network. 4. Benefits on the CDN Similarly, multicast could reduce the downstream bandwidth required on the CDN. The potential for savings follow the same logic as the savings potential on the access network, but instead of analyzing the difference in bandwidth demands for a node, the savings are derived from examining all nodes that a CDN cache is serving, whether it is a region or a hub Society of Cable Telecommunications Engineers, Inc. All rights reserved. 7

8 Table 2 CDN Downstream Traffic CDN Downstream Unicast The number of peak streams per cache population x the average streaming bit rate Multicast The number of unique channels watched per cache population x the average streaming bit rate When this paper is written, CDN caches typically range around Gbps throughput. Shifting Landscape In this section, we will examine specific trends in the industry over the next ten years. We will then explore how these trends line up over time to impact the benefit of multicast. The benefits of multicast are subject to many factors that are changing: 1. Next Generation Access Network reducing the node size and increasing the bandwidth available per subscriber 2. Shifts in viewership and the reduction of linear peak concurrency 3. IP penetration projections and the gradual increase to number of IP-enabled subscribers per node 4. Channel consolidation and the reduction to the number of channels watched per node 5. Uptake of HEVC, lower bit rates and the decrease in bandwidth demand 6. Uptake of UHD and the increase of unique channels viewed and increase in bandwidth per stream 1. Next Generation Access Network As demand for HSD traffic (~50% CAGR) and other unicast video services continues to grow, MSO s are working diligently with industry partners to meet increasing bandwidth demands. We examine the solutions planned and in progress to evolve the network, highlighting the reduction in node sizes and thereby potentially reducing the need for multicast Evolving the Network Operators are executing on several initiatives to increase the capacity of HFC plants, including: Node Splits: Today, a typical node passes 500 households on average, serving 250 subscribers. Splitting network nodes increases the overall capacity per household by decreasing the number of households sharing spectrum. Operators will continue to perform node splits over the next several years. Additionally, Operators will continue to extend fiber past the last amplifier, thus eliminating RF amplifiers, which allows operators to increase usable spectrum beyond current capabilities. This will enable Operators to decrease node size even further. The result is more spectrum shared by less subscribers, providing even more bandwidth per subscriber. DOCSIS 3.1: Deployment of DOCSIS 3.1 allows operators to use wider ranges of spectrum and increases the bandwidth to spectrum ratio (up to ~30%). DOCSIS 3.1 specifications allow up to 2 Gbps upstream 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 8

9 and 10 Gbps downstream assuming 400Hz range at 256QAM modulation and 1.7GHz range for downstream at 1024QAM modulation, a whopping 800% increase over DOCSIS 3.0. In addition, recognizing the bandwidth constraints of HFC, operators are investing in fiber-to-the-home (FTTH), resulting in even fewer subscribers per node: Radio Frequency over glass (RFoG): Substituting the entire coaxial segment of the network with fiber increases the network resiliency. By using RFoG technology, operators can deliver cable services leveraging existing HFC plant components. Passive Optical Networks (PON): Implementation of point-to-multipoint architectures are deployable through either Ethernet networks (EPON) or fiber- (Gigabit-PON) based networks. PON technology significantly increases network capacity. PON is expected to sevice Gbps bandwidth Challenges and Projected Timeline Increasing bandwidth capacity per subscriber will enable operators to offer increased HSD speeds and support more unicast video services, including IP video delivery. However, these HSD initiatives will take some time for operators to implement. For example, the transition to DOSCIS 3.1 alone will require spectrum reclamation and availability, mid-split migration, and the replacement of plant and CPE equipment. Overall, the HFC plant will undergo changes that require significant capital investments and diligent planning. The speed and timing of the plant upgrades and investments, as well as the coordination with other plant initiatives directly influence the opportunity for return on investment. Greenfield and markets with the highest penetration rates will likely benefit from the advantages of these initiatives first. Therefore, the rollout of IP video services may have to follow these time and capital-intensive projects. Otherwise, solutions like multicast will be required to optimize the available bandwidth. Figure 3 summarizes an estimate of the timeline to deploy node splits and PON and the impact they will have on node sizes over the next ten years. Figure 3 Node Size Projections Next Generation Access Network initiatives will decrease subscribers per node and increase bandwidth per subscriber. Both results impact the need and benefit of multicast: 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 9

10 Decrease in node size reduces the chance that multiple subscribers are watching the same channel and therefore reduces the benefit of multicast Increase in bandwidth per subscriber enables increases capacity and therefore reduces the need for multicast 2. Shifts in Viewership Linear video peak concurrency may be the largest driver for multicast. Just like QAM video is broadcast today and optimizes use of spectrum, operators expect that IP video would have to follow suit. However, linear viewership is expected to decrease over time being replaced by other forms of viewing (e.g., VOD, DVR and time-shifting). Peak concurrency refers to a forecast of the peak number of subscribers that will be watching any content at the same time. Where there is high peak concurrency, there is both a high network bandwidth requirement for delivery and, by virtue of the law of large numbers, an improved probability that subscribers will be watching the same content at the same time. In a survey by Hub Research, we see that the majority of viewers now suggest their default option is timeshifted TV viewing. See Figure 4. Figure 4 Live TV as Default TV Source Where there is high peak concurrency within large nodes, it presents an opportunity to achieve network optimization benefits with multicast delivery. However, trends suggest the opposite: As described with Next Generation Access Network, node sizes are decreasing At the same time, peak concurrency for linear viewership are also decreasing We predict that linear viewership will continue to decline and that all forms of time-shifted viewing will continue to climb. See Figure Society of Cable Telecommunications Engineers, Inc. All rights reserved. 10

11 Figure 5 Peak Concurrency Projections Reduction in linear viewership will result in a reduced linear peak concurrency. A reduction in peak linear concurrency will reduce the probability that more than one household watches the same linear channel at the same time thus reducing the benefit of multicast on the access network. 3. IP Penetration Historically, operators have experienced that major STB transitions take time (e.g., analog to digital, MPEG2 to MPEG4, etc.). While operators are working to overcome technical challenges, rollout of all IP STBs will not happen overnight. Roadmaps from large operators suggest IP STB rollouts of 1%-5% penetration in year one and eventual ramp-up to 8%-20% year-over-year penetration. Operators expect it will take up to 10 years to transition to all IP STBs, with target dates well into the mid-2020s. See Figure 6. Figure 6 IP Penetration Projections Typically, operators find it is most cost effective to roll out new STBs without forced truck rolls. We call this strategy cap and grow, where STBs are replaced only for new subscribers, moves, break-fix or any other needs for a technician to go to a subscriber home. Alternatively, operators could rip and replace existing STBs with additional truck rolls to force a higher IP video penetration more quickly. However, this is costly and not a strategy many operators follow. Assuming operators roll out IP STBs at a pace similar to previous STB roll outs, IP penetration may not reach enough scale for multicast to add a benefit Society of Cable Telecommunications Engineers, Inc. All rights reserved. 11

12 4. Channel Consolidation The number of available linear channels in a lineup impacts the benefit of multicast. When there are very few channels, the likelihood of multiple subscribers watching the same channel are increased, and vice versa. While not intuitive, there is reason to think the number of channels will both grow and shrink. The arguments for a growing number of channels include: IP networks enable near seamless additions of linear channels Barriers to content creation are now very low and the sheer amount of content available to become a channel is increasing Content holder rights contracts may hinder any consolidation in the near-term and as operators go IP, more channels may be added in the interim The arguments for a shrinking number of channels include: Subscribers already believe there is limited quality content spread across too many linear channels Subscriber behavior is trending away from linear viewing to time-shifted alternatives and overthe-top (OTT) content Government regulations or operator strategies may evolve towards an a-la-carte option Given the above, we are likely to see a near-term expansion that may or may not be followed by widescale channel consolidation. Operators must plan for either scenario. A significant shift in the number of channels impacts several intertwined areas, including cost and network demand, subscriber behavior, revenue, product offering and broadcaster strategy. This paper focuses specifically on the the benefits that multicast brings. The impact would depend on the degree of consolidation or growth, but it stands clear that with less linear channels, more subsribers will be watching the same channel at peak on average. Therefore, the projected efficiencies of multicast will grow as channels shrink. The opposing story with more channels added to the lineup would result in themulticast benefit decreasing as viewing becomes more fragmented. Two examples shed further light on this dynamic: 1. If the industry consolidated down to 20 channels, operators could efficiently multicast every channel in every node. The total linear DOCSIS traffic would be 20 streams in the various required formats. 2. If there are thousands of channels, concentration of peak linear viewing would be too thin for all but the most highly watched events. Operators would have to determine the multicast benefit for the Super Bowl, not an average peak linear viewership night. 5. Uptake of HEVC The amount of bandwidth required to deliver IP video over the access network impacts the benefit of multicast. HEVC aims to decrease the bit rate requiring less bandwidth per subscriber than with MPEG2 or MPEG4 (AVC) for HD and UHD content Society of Cable Telecommunications Engineers, Inc. All rights reserved. 12

13 The uptake of HEVC is tied to the support of encoders (operator network) and decoders (STB in subscriber homes) and other barriers such as licensure. To take advantage of improved compression, the STB or IP Player must be able to decode HEVC. Operators are starting to deploy HEVC capable STBs, and investing in HEVC transcoding to support 4K and HDR content and to reduce bit rates for HD content. Consumer devices are also now HEVC-capable. HEVC is expected have up to a 50% bandwidth improvement (lower bit rate) over MPEG4 AVC, which could significantly reduce the need for multicast. However, the rollout of HEVC will take time. Until all subscribers have HEVC-capable devices, operators must support both AVC and HEVC for IP video delivery. To optimize video delivery, operators will likely choose to multicast the least common format (AVC) or unicast the most efficient format (HEVC) when possible. As more devices support HEVC, the benefit of multicast is reduced since less bandwidth will be required to deliver IP video encoded with HEVC. Once HEVC compatible IP STB are deployed, they will follow the IP STB penetration trends. Since IP STB penetration is estimated at 8%-20% YOY, we can expect the same 8%-20% YOY HEVC penetration. 6. Uptake of 4K and HDR The amount of bandwidth required to deliver IP video over the access network impacts the benefit of multicast. 4K and HDR content will increase bit rates, requiring more bandwidth per subscriber than HD content. HEVC reduces the bandwidth required for IP video streaming. However, the improvements may not be felt by operators due to the increase in bit rates to support 4K video and the use of HDR. 4K is estimated to require around Mbps in HEVC per stream. HDR is estimated to increase bit rates (either HD or 4k) by up to 10%. Similar to HEVC, 4K and HDR must be supported by the operator s video processing system and the IP Player in the home. There are two key impacts of 4K and HDR video: 1. Increased bit rates requiring more bandwidth to deliver video to 4K/HDR capable devices 2. More diversity in unique channels watched since the HD version of a channel is a different video stream than the 4K/HDR version(s) of the same channel The result is that 4K/HDR streams are incremental to the channel line-up and segment subscribers within a node. For example, assume a node where the top 5 channels are available in both HDR and HD. If both the HDR and HD are watched, those 5 unique channels require 10 unique streams. The result is a decreased benefit of multicast on the access network since it creates more unique channels viewed at peak. Unlike HEVC, operators cannot simply multicast the least common format since subscribers likely are paying for a 4K/HDR feature Putting it all Together Any one of the shifting trends could make a case for or against multicast if quickly taken independently and to an extreme. However, we know that many of the trends will happen gradually and together. For 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 13

14 example, over several years, Next Generation Access Network enhancements will make way for more IP STB penetration, during which linear viewership is likely to decline. Figure 7 Factor Overlay If an operator were to transition to 100% IP STB penetration overnight given today s network capacity and viewership dynamics, then multicast would more than likely be required and offer significant benefit. However, the emerging hypothesis is that given how these trends will play out together and over time, multicast may not offer the benefit once anticipated over thelong term. Multicast, however, may play a pivotal role during the transition from QAM to IP video delivery. 7. Likely Senarios and Impact to the Network Based on industry averages, we investigate three likely reference scenarios for short-term, mid-term and long-term to compare CDN demand, access network downstream demand and access network upstream demand. We also investigate a base case that represents how 100% IP penetration would impact the demand on the network if all other levers are held constant to today s landscape. The remaining scenarios for short-term, mid-term and long term will apply changes in landscape to analyze the impact of multicast with shifting variables. In each of the scenarios, we also examine the impact with IP local DVR. If recordings are stored in the cloud, record concurrency will not impact bandwidth requirements on the access network. Record concurrency is defined as the peak simultaneous recordings per household. The playback from the cloud for cloud DVR will impact bandwidth on the access network, but this traffic will all be unicast Society of Cable Telecommunications Engineers, Inc. All rights reserved. 14

15 Table 3 Scenario Assumptions Base Case Short-Term Mid-Term Long-Term Peak Linear Concurrency 80% 80% 65% 45% Peak Record Concurrency 125% 125% 140% 175% IP Penetration 100% 25% 50% 75% Node Size Average Streaming Bit Rate Unique Channels Watched at Peak Per Node Unique Channels Watched at Peak CDN Population Unique Channels Recorded or watched at Peak per Node Unique Channels Recorded or watched at Peak CDN Population Subs per CDN Population HD Channel Line Up DVR Take Rate 65% 65% 68% 70% 7.1. Base Case In the base case, we use a snapshot of today s node size, linear concurrency and average streaming bit rate combined with the future 100% IP penetration. As expected, if the 200 streams at peak per node (250 subscribers x 80% peak linear concurrency) were concentrated to 100 unique channels, an MSO could realize 50% bandwidth efficiencies by deploying multicast in the access network. Table 4 Base Case Scenario Without Local DVR: Multicast Unicast Delta Delta in Estimated Units CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per node (Mbps) With Local DVR: DOCSIS 3.0 Channels CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per Node (Mbps) 563 1,814 1, DOCSIS 3.0 Channels 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 15

16 7.2. Short-term Scenario The short-term scenario is likely in the next few years, with larger node, high linear peak concurrency, 25% IP STB penetration and AVC delivery with some HDR. In this scenario, the maximum peak streams per node are 50 streams (25% STB penetration, 80% concurrency and 250 subscribers per node). As shown below, in the case without local DVR, multicast has the potential to save 80 Mbps per node, the equivalent of approximately two DOCSIS 3.1 channels. Local DVR increases the benefit of multicast. In this example with 125% peak record concurrency and a 65% take rate multicast reduces access network bandwidth required by 223 Mbps per node. Table 5 Short-term Scenario Without Local DVR: Multicast Unicast Delta Delta in Estimated Units CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per Node (Mbps) With Local DVR: DOCSIS 3.1 Channels CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per Node (Mbps) DOCSIS 3.1 Channels In this example, the maximum peak streams are 50. If 50 unique channels are watched at peak, the benefit of multicast on the access network drops to 0. Figure 8 illustrates the sensitivity of bandwidth savings for a range of unique channels watched Society of Cable Telecommunications Engineers, Inc. All rights reserved. 16

17 Figure 8 Access Network Bandwidth Required Per Number or Unique Channels Watched 7.3. Mid-term Scenario The mid-term represents a likely scenario in roughly five years. In this scenario, we are comparing smaller nodes from node splits, HEVC, some 4K and HDR, 50% IP penetration and decrease in linear peak concurrency to 65%. In this scenario, the maximum amount of peak streams is approximately 49. Even without DVR, this scenario shows a savings of 79 Mbps on the access network. On local DVR, this scenario shows a savings of 309 Mbps. Since the maximum streams in this scenario is approximately 49 streams (150 subscribers per node x 50% IP penetration x 65% peak linear concurrency), the sensitivity analysis from the short-term analysis is largely applicable. Table 6 Mid-term Scenario Without Local DVR: Multicast Unicast Delta Delta in Estimated Units CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per node (Mbps) With Local DVR: DOCSIS 3.1 Channels CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per Node (Mbps) 7.4. Long-term Scenario DOCSIS 3.1 Channels The long-term represents a likely scenario in approximately seven to eight years, capturing smaller nodes from FTTH and a continual decrease in linear concurrency. The savings in this scenario are less than in the previous scenarios, even though the IP penetration is increased. Without local DVR, this scenario shows a savings of 14 Mbps. Withlocal DVR, the savings is 153 Mbps per node on the access network. In this scenario of 40 subscribers per node, 45% peak linear concurrency and 75% IP penetration, the maximum number of streams is Table 7 Long-term Scenario Without Local DVR: Multicast Unicast Delta Delta in Estimated Units CDN Sizing (Gbps) Gbps Caches 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 17

18 Access Network downstream per node (Mbps) With Local DVR: DOCSIS 3.1 Channels CDN Sizing (Gbps) Gbps Caches Access Network Downstream Per Node (Mbps) 8. Costs to Deploy Multicast DOCSIS 3.1 Channels If deploying multicast was free and easy, then any savings could argue the deployment of multicast. Deploying multicast consists of several costs. First, the MSO has to choose a solution. Figure 9 Deployment Scenarios To calculate the costs of deploying either a multicast or unicast solution, the location of the gear must be considered. The number of servers and caches are based on the number of points of presence (PoPs) and the number of subscribers they support. If the servers are deployed lower in the network, there will be 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 18

19 more servers and, therefore, more costs. Placing a cache lower will reduce the network traffic costs of certain links, but increase the number of caches required. In scenario one depicted in figure 9, by placing the multicast server north of the hub router, the traffic between the metro router and the hub router will equal the number of unique channels watched at peak x bit rate. Even though scenario two uses multicast, because the multicast server is south of the hub router all of the traffic between the metro router and the hub will be unicast and the same as in scenario three. The traffic in scenario two and three will also equal the number of unique channels watched x bit rate. The key difference in the video distribution/cdn is the requests and responses to the edge cache. In scenario one and two, only unique channels are streamed from the edge cache. In the case of unicast, all streams are delivered from the edge cache. Caches typically scale by storage and network throughput, with the least common denominator driving incremental hardware needs. Typically, the network throughput is the bottleneck. Cache throughput varies per solution, with a sample cache potentially having Gbps throughput. The other costs to consider are in the Multicast Client. Aside from licensing costs for the Multicast Client, there is the additional memory required to cache the multicast data. While this may be small per client, there are large numbers of clients. Lastly the operator will have to operate and maintain the solition. Conclusion Early on, MSOs were confident that multicast was needed to transition to IP video. There are a lot dynamic developments that shift the equation for the expected benefit of multicast. MSOs must closely examine an aggregate of their own roadmaps of IP penetration, DVR architecture, access network Initiatives and rollout of HEVC, 4K and HDR to determine if multicast is needed for them. For MSOs with small nodes and low streaming bit rates, multicast may not be worth the complexity. Abbreviations ABR AVC BW CAGR CDN CPE DASH DOCSIS DVR EPON FTTH Adaptive Bit Rate Advanced Video Coding Bandwidth Compound Annual Growth Rate Content Delivery Network Customer Premise Equipment Dynamic Adaptive Streaming over HTTP Data Over Cable Service Interface Specification Digital Video Recording Ethernet Networks Fiber-To-The-Home 2016 Society of Cable Telecommunications Engineers, Inc. All rights reserved. 19

20 Gbps Gigabits Per Second Gigabit-PON Gigabit Passive Optical Network HD High Definition HDR High Dynamic Range HEVC High Efficiency Video Coding HFC Hybrid Fiber-Coaxial HLS HTTP Live Streaming HSD High Speed Data HTTP HyperText Transfer Protocol IGMP Internet Group Multicast Protocol IP Internet Protocol mabr Multicast ABR Mbps Megabits Per Second MPEG2 Moving Picture Expert Group 2 MPEG4 Moving Picture Expert Group 4 OTT Over-The-Top PON Passive Optical Networks POPS Points of Presence QAM Quadrature Amplitude Modulation RF Radio Frequency RFoG Radio Frequency Over Glass SDV Switched Digital Video STB Set-Top Box VOD Video On Demand UHD Ultra High Definition Bibliography & References Hub research: What s TV Worth, April 2016 Cable Labs Version 01. IP Multicast Adaptive Bit Rate Architecture Technical Report Society of Cable Telecommunications Engineers, Inc. All rights reserved. 20

21 Huang, Maxwell. HFC Improvement For DOCSIS 3.1 Evolution Spring Technical Forum Proceedings Society of Cable Telecommunications Engineers, Inc. All rights reserved. 21