Watch Telephony's Security Webcast Find out what Verizon Business is doing with managed network security and how service providers can cash in with this new service. Watch Telephony's Webcast today.

As incumbents face increasing competition from cable operators, they are countering by increasing DSL speeds, accelerating fiber deployments, and preparing to roll out advanced IP services such as video on demand (VoD) and IPTV. However, increasing per-subscriber bandwidth and assuring quality of service (QoS) for video offerings requires incumbents to make critical decisions about designing their access networks for optimization of capex and opex resources. Many are struggling with the fundamental decision of where to place the critical subscriber management features and IP routing intelligence: should they be consolidated in a small number of centralized point of presences (PoPs) or distributed closer to the consumer using IP-enabled DSLAMs?

The deployment of video services over DSL presents an interesting challenge for network operators that need to understand the total cost of ownership (TCO) of their access infrastructure. Architectural considerations must bring into account the network at hand--while some considerations may be true for a specific network, they may be altogether wrong for another. The basic choice is between centralized or distributed subscriber management.

Not long ago, many were touting distributed subscriber management, because IP-enabled DSLAMs could more efficiently handle multicast, replicating traffic streams as close to the target as possible. However, scattered attempts to deploy this architecture have revealed that it creates more problems than it solves, because any gains from applying the intelligence at the edge are consistently offset from the difficulties of managing this architecture.

Proponents of distributing subscriber intelligence to the edges of the access network claim that centralization introduces a single point of failure for all subscribers. However, this argument is based upon the false premise that there is a single large broadband services router (BSR) in which all traffic passes. Real-world centralized architectures usually utilize a handful of PoPs that act as aggregation sites for specific geographical areas. This means that there is no “single point of failure”--in fact, the centralized access architecture increases the operational efficiency and speed in which network maintenance and trouble shooting can occur. A reduced number of IP-aware platforms in a few PoPs are far easier to support than a large number of devices spread throughout central offices (COs) and possibly neighborhoods.

In smaller networks the architecture question will not be raised: a handful of BSRs with subscriber management and QoS is the most efficient choice. Even larger networks with millions of subscribers can be easily managed with closer to fifty broadband routers, centralized in a small number of PoPs. Each BSR aggregates hundreds of DSLAMs and thousands of subscribers.

Alternatively, distributing subscriber intelligent to the DSLAM would mean that the number of IP-aware platforms would increase from the 50 or so centralized BSRs to thousands--or tens of thousands. The expenses associated with procuring, servicing, troubleshooting and rolling out software upgrades to possibly tens of thousands of IP DSLAMs makes the distributed architecture unrealistic.

Common sense indicates that the network architecture with fewer IP addresses and fewer devices to purchase, troubleshoot and manage will save service providers both capex and opex and help them to become profitable faster.

In the case of DSL access networks, it is in the service provider’s best interest to keep the most commonly found equipment commoditized. And because the DSLAM is the most common equipment, and constitutes the vast majority of capex/opex, it should be the least expensive and easiest to manage. This is quite the opposite of what we find in IP DSLAM product development. In order to increase value in the eyes of the service provider, many vendors have product development programs that are aimed at migrating IP processing and subscriber management into the DSLAM. However, the addition of network processors, ASICs, FPGAs, memory, routing code, PPP, DHCP, OAM, QoS, control and manageability will all increase the costs. These additional costs, plus the cost to maintain and upgrade DSLAMs, will be quite burdensome for service providers. Supporting 40+ broadband services routers is a much more attractive option that supporting thousands of IP DSLAMs.

A centralized access architecture also offers the ability to effectively support a range of access architectures in addition to DSL. Centralized locations can efficiently aggregate, and offer subscriber management for Ethernet access, PON and 802.11 wireless services as well as DSL. This allows service providers to share network resources such as RADIUS, accounting, address management and Lawful Intercept across all access technologies, reducing duplication of resources and lowering both capex and opex.

In addition to locking out the use of other access technologies, one of the largest drawbacks to the distributed architecture is that it can easily lock service providers into a single-vendor solution for the major cost component of the network. Either the IP card will be located on the remote DSLAM or there will be a proprietary technology used for the remote DSLAM to communicate with the IP processes in a larger IP DSLAM controller unit. In any event, this architecture locks service providers into a single solution when control functions are on the DSLAM. Service providers should be under constant investigation to lower their costs. As new technologies emerge (e.g., WiMAX), service providers should be able to take advantage of lower-cost venues to service their end-users.

While distributed architectures require significant changes to service providers’ provisioning models, the centralized model allows service providers to maintain current provisioning models. This is most evident in a multi VC or VLAN architecture where the number of VCs per subscriber is equal to the number of services each subscriber requests. The increased number of VCs or VLANs per subscriber requires massive provisioning of IP DSLAMs and other network devices to support the additional VCs and/or VLANs.

In order to avoid the added expense and burdensome configuration of this architecture, the DSL Forum published the TR-059 recommendation for the support of QoS-enabled services. TR-059 specifies that broadband subscriber management should not be located in the DSLAM and that a single VC per subscriber should be provisioned to deliver all of the services that the end user has subscribed to.

According to TR-059, subscriber management should be located as close as possible to the DSLAM in order to provide better control of customer traffic, and to manage the traffic as close as possible to its source. Each aggregation point between the traffic source and the point of subscriber management degrade QoS. For most networks, selecting a centralized architecture provides all the advantages of reduced cost and ease of management while maintaining control over the customer traffic.

More recently, the question has been raised whether current generations of BSRs will become overburdened as DSL speeds increase and services like video on demand become more prevalent. While the question of the impact of IP video on network architectures raises interesting points, it still does not change the fundamental capex, management, and opex advantages a centralized architectural approach provides. In the end, the question is not whether video and other applications will drive a distributed approach to IP intelligence, but how it will drive BSRs to evolve to meet these challenges: with increased capacity and density, with increasing levels of service awareness, or with some combination of these and other capabilities.

As service providers plan the rollout of triple play services, it is critical that they understand how different architectures will affect their capex and opex. Obviously, a simple architecture is always better from a cost-savings perspective. Service providers, however, should not only look at the BSR-related capital in the architecture, but how a certain implementation of a platform can lock (or not lock) a provider into a single vendor solution.

A more centralized approach provides unparalleled simplicity and efficiency that reduces the number of IP-enabled devices needed to operate and manage a triple play access network. It also delivers on reducing capex and opex, a key factor when service providers evaluate whether to incorporate a new technology or vendor into an established network.

Dave Boland is Juniper's senior product marketing manager.

Visit Juniper Networks online.