Impacts of Over-Provisioning in VOD Transport Networks
Does the "cost-per-stream" metric underestimate the true price tag of a provisioned subscriber? Can a Layer 2 distributed Ethernet switching solution minimize your costs and offer improved flexibility?
Among the greatest challenges posed by video-on-demand (VOD) is that of building a cost-effective, flexible and scalable network for the transport of VOD traffic. This issue is particularly challenging in light of the fundamentally different nature of VOD traffic when compared against other services that have been deployed across the existing HFC infrastructure.
With the introduction of gigabit Ethernet (GigE) as a new means of transport, and the replacement of the legacy digital video broadcast (DVB) asynchronous serial interface (ASI) transport technology, the engineering community has achieved a major improvement. However, most current Ethernet implementations (Layer 1, or "point-to-point") actually lead to an underutilization of your network assets. Furthermore, this shortcoming is not adequately reflected in the "cost-per-stream" metric that is commonly used to summarize the economics of VOD transport networks today.
VOD represents a substantial opportunity to generate incremental revenues, while also controlling churn or the loss of subscribers to direct broadcast satellite (DBS). However, VOD and other interactive video services are very different than services deployed previously by cable operators.
VOD in traditional networks
Traditional cable networks are broadcast-centric, that is, they are suited to provide "one to many" (1:N) services. With the introduction of pay-per-view (PPV), this same networking paradigm continued, albeit with interdiction to provide limited access and enable billing against a premium service.
The advent of the cable modem provided a new and challenging networking architecture for cable operators. However, Data Over Cable Service Interface Specification (DOCSIS) standards for interconnection with the HFC plant, in combination with traditional data networking architectures for transporting traffic around the metro area, provided a proven solution to support cable modem traffic.
While cable modem service appears to be fundamentally different from broadcast and PPV services, there are in fact many similarities. Cable modem traffic also is broadcast to all participants on a given segment, although the traffic is captured by the cable modem only to which it’s addressed. As a result, to date, all of the new services being offered by cable operators across existing HFC networks have matched this broadcast paradigm exceptionally well.
VOD is fundamentally different than other services because it is a "one-to-one" (1:1) service. This means that there is a single video stream being dedicated to a single customer. To understand the full impact this service has on network and operations, it must be deconstructed into its two basic parts, which are shown in Figure 1 the HFC and VOD transport portions of the network.
VOD in the HFC portion
The HFC portion of the network for a VOD service is strikingly similar to the one for PPV service. A video stream is broadcast across the HFC network and the set-top box provides the interdiction and control to limit access to this channel.
However, unlike PPV, a VOD channel is being viewed by only a single customer at any given point in time. This difference requires you to provide a number of channels for VOD services within each HFC segment that is sufficient to satisfy the maximum number of subscribers that may request service at any one time ("peak demand"). These channels will be specifically reserved for on-demand services and will not be available for other services, thus representing a committed resource.
Subscribers from many HFC segments are typically served by a single VOD server. Because there are variations in demand across HFC segments, the VOD server is often sized to serve a fewer number of VOD streams than the sum of the peak demand from each segment.
This approach ensures that expensive VOD server resources can be optimized such that the probability that a subscriber is denied service is kept below an acceptable level. As a consequence, the total number of channels reserved within the HFC segments across the entire network is typically larger than the number of video streams available from the VOD server.
The ratio of the number of channels reserved within the HFC segments to the number of VOD streams available on the VOD server represents the "over-provisioning ratio" of the VOD network. In typical networks today, these ratios run from as low as 2 to as high as 5. While this level of over-provisioning will, by definition, result in overall under-utilization of the HFC network infrastructure, operators consider it to be the necessary cost of providing a 1:1 service across a broadcast medium. Fortunately, the benefits to on-demand service subscribers far outweigh the cost of over-provisioning in this portion of the network.
VOD in the transport net
The next portion of the network we’ll consider is the transport network from the VOD server to the quadrature amplitude modulation (QAM) insertion point, or the transport network. This is the network dedicated to the distribution of VOD traffic. It includes the fiber transmission systems as well as the switching system(s) required to distribute VOD streams from the VOD server to the QAM insertion point. It is this portion of the network that provides the greatest opportunity for substantial CAPEX (capital expenditure) and OPEX (operating expense) savings.
Early VOD implementations have carried the broadcast paradigm of the HFC network through the transport network and, in many cases, into the VOD server. Such implementations are typically based on DVB-ASI, point-to-point Layer 1 GigE, or in some cases telecom carrier-centric Layer 2/3 solutions.
Layer 1 GigE economics
As we’ve said, one of the major, and nearly overlooked, characteristics of any point-to-point GigE solution is that over-provisioning is required in the transport network.
Figure 2 presents an example to illustrate the issue you face when deploying Layer 1 ("point-to-point") transport solutions. In the two-hub example shown, the HFC networks fed from the hubs can support peak demands of 3,000 streams each. (The peak demand—3,000 streams in this example—in each HFC network segment is calculated as the product of the total VOD-enabled subscribers and the concurrency ratio, the percentage of VOD-enabled subscribers demanding content simultaneously.)
Because the actual demand between the two hubs differs at any given time, a VOD server with a capacity of 4,000 streams is considered adequate to serve the aggregate demand of the two hubs to keep the probability of a subscriber being blocked below an acceptable level.
Because no switching capability is available at either hub, an aggregate of 6,000 streams of network transport (that is, 3,000 streams from two hubs) must be deployed between the hubs and the headend. A Layer 2 switch at the headend then performs the task of switching up to 4,000 streams from the VOD server to any of the 6,000 streams of transport that carry video-streams to the hubs. Because the VOD server can never deliver more than 4,000 streams at any time and the transport network has a capacity of 6,000 streams, the network is over-provisioned by a factor of 1.5. In real-world scenarios, this factor typically ranges from 2 to 5.
This simple example demonstrates why cost-per-stream does not accurately reflect your true CAPEX impact. You’re required to build a network with a capacity of 6,000 streams, while the VOD server’s capacity is only 4,000 streams (that is, the number of "provisioned subscribers" is 4,000). Or, in other words, if the assumed cost-per-stream is $30, the transport network cost is 6,000 x $30 = $180,000, and the "cost-per-provisioned-subscriber" is actually $180,000/4,000 = $45.
So, cost-per-stream underestimates the true cost for a provisioned subscriber. A precise definition of cost-per-provisioned-subscriber is the ratio of the cost of a transport network that can deliver the number of streams equal to the capacity of a video server, and the capacity of the video server measured in number of streams.
Dynamic bandwidth provisioning
Alternative approaches exist that result in better economics for the transport network. In one solution, the transport network is integrated with a switching platform. This approach, commonly referred to as Layer 2 distributed Ethernet switching, enables you to implement a VOD transport network that matches the VOD server while being able to deliver VOD streams to a much larger number of QAM channels via the HFC network.
By introducing switching capability closer to the customer, you can achieve cost savings while improving the customer’s experience. This bandwidth-on-demand architecture enables the dynamic allocation of network bandwidth and provides a dynamic, on-demand (1:1) centric network. This approach, unlike the broadcast-centric alternative, recognizes the 1:1 nature of on-demand services.
Figure 3 shows a Layer 2 solution for VOD transport networks. Because streams can be switched by the network dynamically, the transport network required can now be limited to the capacity of the VOD server, say 4,000 streams. Note that each hub still supports a peak demand of 3,000 streams without requiring any over provisioning in the network.
With a $30 cost-per-stream (as in the Layer 1 GigE example), the total transport network cost would be 4,000 x $30 = $120,000, and because the number of provisioned subscribers is still 4,000 (capacity of the VOD server), the cost-per-provisioned-subscriber also is $120,000/4,000 = $30. Clearly, the cost-per-provisioned-subscriber metric is more accurate in terms of quantifying the total CAPEX required to build the transport network.
Other benefits
In addition to the CAPEX cost savings, a Layer 2 solution for VOD transport networks has several other advantages, which include:
OPEX savings because of improved management capabilities, and the fact that a lower amount of network equipment is required.
Choice of centralized, distributed or hybrid architectures. For example, a "Top 50 server" can be deployed earlier and be utilized more efficiently because of switching capability available at hubs.
High reliability because of the ring architecture (logical ring with physical ring or star) that can survive fiber cuts.
Pay-as-you-grow scalability.
The 1:1 nature of on-demand services requires a networking understructure that is 1:1-centric. Only by deploying a network that integrates network transport with a distributed switching solution while simultaneously minimizing CAPEX can you achieve the true benefits of on-demand services.
Greg Cinque is vice president of engineering and Jan Mul is vice president of marketing at Manticom Networks. Reach them at [email protected] and [email protected].
Did this article help you? Send email to [email protected].
DVB-ASI: This is the legacy transport protocol. Because this protocol was basically built for video broadcast, it suited this application very well and was highly cost-effective to deploy. However, when required to support an interactive service such as VOD on a large scale, it is expensive to maintain and its scalability is limited.
Telco Carrier Centric Layer 2 and Layer 3 Solutions: These solutions, based upon resilient packet ring (RPR) or packet-over-SONET (POS), provide the required flexibility and scalability. However, most Layer 2 solutions are built for the telco carrier market and have prohibitive costs because of the additional features and functions required for that market. Similarly, Layer 3 solutions have a cost structure that makes them not competitive for the cable VOD market.
Layer 1 Gigabit Ethernet: At first glance, current Layer 1, or point-to-point, GigE transport solutions appear to be the best answer for VOD transport networks. They are cost-effective, that is they provide a low cost-per-stream, a popular metric used in the industry. A second look, however, reveals shortcomings given the nature of the VOD range of services:
1) Point-to-point (Layer 1) GigE architectures clearly favor centralized video server architectures, thus leaving the cable operator with no flexibility in deploying either a decentralized or centralized video server solution.
2) Typically, Layer 1 GigE solutions provide only limited network management capabilities. Therefore, the visibility of the network is limited and troubleshooting is simply not possible prior to the inevitable phone call that a subscriber makes when he/she has a complaint about the service.
3) Granularity limitations are an issue in smaller-scale deployments. Some of the existing solutions can be implemented only in 10 Gbps increments, which may be much more than the typical cable operator requires.
4) Over-provisioning of bandwidth also is required in the transport section of the network. This is because Layer 1 solutions do not provide any switching capability. So, peak bandwidth must be transported to each HFC segment. Unlike with the HFC section, however, asset utilization, or lack thereof, becomes a real issue. This issue seems to be somewhat overlooked as a factor in the early deployment of VOD services.
VOD in a One-on-One World
BOTTOMLINE
On-demand services are fundamentally different in nature than the services you’ve historically offered across your HFC networks. While broadcast, PPV and cable modem services are very different in the experiences they deliver, they are all a broadcast-centric, or "one-to-many," services.
On-demand services, in contrast, are "one-to-one" services. This fundamental difference in service type requires that you consider the provisioning and deployment of on-demand networks from a substantially different viewpoint. While early implementations have utilized broadcast-centric networking models, matching the size of the on-demand transport network to the size of the QAM insertion, this approach is expensive and is counterproductive to the very nature of on-demand services.
By moving the point of dynamism as close as possible to the customer (the QAM insertion point), you can achieve substantial cost savings over broadcast-centric solutions, whether they are based on DVB-ASI or GigE. In addition, a dynamic, one-to-one centric network offers additional benefits in the form of much greater flexibility in the deployment of Top 50 servers and a multitude of options in growing on-demand services.
