Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Networks

Overview: Distributed IP/Ethernet DSLAMs

Posted: 01 Mar 2006 ?? ?Print Version ?Bookmark and Share

Keywords:Sachin Kumar Jain?

By Sachin Kumar Jain

With growing complexity and increased packet processing and throughput requirements, there is a greater need for a cost effective distributed DSLAM architecture so that traffic-processing requirements are performed in the first mile. With the evolution of new standards and their increased capacity to compete with existing ones, these requirements are being addressed.

Next-generation DSLAM architectures in access networks require traffic processing devices to address a wide range of feature requirements. And, cost points must be established that enable the advantages of a distributed pay-as-you-grow equipment and network architecture.

In the past, DSLAMs have used ATM on both the network and customer sides. More recently, there is a trend towards Ethernet or IP-based network links using, for example, Gigabit Ethernet connections.

Next-generation architectures
ATM provided a QoS-capable technology that could extend from the subscriber all the way to the ISP. In next-generation architectures the DSLAM will terminate the DSL subscriber's ATM circuit and must convert to a packet-based technology that reliably and securely delivers subscriber traffic to the service provider. Telecom service providers worldwide are defining and driving standardization of new network architectures. These architectures must support value-added services that enable telecom service providers to compete with such broadband service providers as cable TV companies, as the telecom service providers drive become the provider of choice for the "triple play" of voice, data and video services.

A primary requirement of new network architectures is the ability to support IP services natively to take advantage of the rich service offerings that IP enables, and the use of low-cost Ethernet transport and switching technologies. This led to a new class of next-generation DSLAMs based on packet technology rather than ATM.

A typical packet-based DSLAM architecture is based on Ethernet "Layer 2" DSLAMs. For these DSLAMs, subscriber traffic is backhauled to a Metro Broadband Remote Access Server (BRAS), also called a Broadband Network Gateway (BBNG), using Ethernet VLANs in combination with Ethernet Multicast capability. This allows the backhaul network to be based on relatively inexpensive Ethernet networks, compared to ATM. In some cases IP "Layer 3" functions will be added to DSLAMs, ranging from BRAS off-load capability (e.g. local session termination/aggregation) for smaller remote DSLAMs, to a full-blown BRAS capability for large, centralized DSLAMs.

In yet other situations, Multiprotocol Layer Service may be used to transport traffic through the regional broadband network. Next-generation DSLAMs must provide an architecture that can simply and cost-effectively adapt to any of these scenarios.

Legacy DSLAM architecture
A typical DSL network architecture deployed during the initial rollout of ADSL (1999-2001) used an ATM PVC from the customer to the BRAS, at which point PPP sessions were either terminated or tunneled to the ISP using L2TP. The BRAS performs session management (AAA functions and IP services) and acts as an interface to the network service provider (NSP).

In such architecture, the DSLAM serves as a simple Layer-2 ATM multiplexer or concentrator. For ATM-based DSLAM applications, ATM only devices and architectures have proven to be successful solutions. However, the DSLAM market sees a rapid introduction of Ethernet and IP DSLAMs. These types of DSLAMs require ATM traffic processing for traffic over the DSL loop and packet processing capability for backhaul network connections.

Figure 1: Legacy DSLAM Architecture

DSLAM architecture evolution
The force behind the evolution of the DSL access network from ATM to Ethernet is two fold. First, network architectures must move far beyond the current architecture that supports only best effort traffic to a multi-service architecture supporting a rich mix of enhanced, "triple-play" services enabled by advanced traffic processing and quality-of-service (QoS) mechanisms.

Second, ATM-based access networks, although capable of supporting these types of services, do not scale well and are too expensive to provide an acceptable solution. This caused vendors to look at Ethernet and IP-based architectures to address the shortcoming.

The evolution of Ethernet based architecture can be shown in stages (See Figure 2). The first step was a traditional broadband architecture with any of the DSL combinations (ADSL, SDSL, and VDSL) delivered by ATM over DSL in the first mile. In the second mile, from the DSLAM upwards, it is transported by a variety of ATM variants up to an ATM switch then into a BRAS.

The next step involves taking advantage of Gigabit Ethernet, a technology that did not exist when the traditional broadband architectures were set in the mid-1990s. Using Gigabit Ethernet in this way reduces interface costs in the DSLAM itself and also in the aggregation of traffic in the Ethernet switch and into the BRAS.

The final step is an evolution towards an EFM (Ethernet in the first mile) standard. This occurs when fiber is brought deeper into the access network, either to a node supporting EFM copper (Ethernet over VDSL) or with a fiber solution (EFM over single mode fiber).

Figure 2: DSLAM Architecture Evolution

Ethernet based DSLAM architecture
One way to conceptualize a next-generation packet-based DSLAM architecture is to consider the basic operations on subscriber traffic as it flows into the network. First it is processed by an ATM AAL5 SAR to convert ATM-based DSL traffic to Ethernet frames. Then, the Ethernet traffic is switched to an aggregation card to be transported to the network. Along the way, a number of operations are performed on the frames.

A great feature for supporters of Ethernet for DSL backhaul is that it forms part of a bigger picture of a public Ethernet evolution. This builds on the technology advances in metro Ethernet from the 10 Gigabit Ethernet Alliance (10GEA) initially, and then through the current work of the Metro Ethernet Forum and the developments for Ethernet in the first mile (EFM). An overview of the public Ethernet architecture follows:

The basic argument for a public Ethernet approach is that it offers Ethernet-based solutions, which:
? Provide flexibility
? Provide scalability
? Offers low equipment cost
? Meets short-term needs for bandwidth in the second-mile network.
? Provides operators with a technology platform that can be migrated into the first mile (between the subscriber and the first network node) as bandwidth requirements increase in the future, both over copper and fiber deployments.
? Reduces cost. A strong argument in favor of Ethernet upstream of the DSLAM is based on the fact that capital expenditure (capex) is reduced because the expensive high-capacity ATM switches are eliminated, a crucial fact when the number of ATM PVCs required by broadband services is increasing rapidly.
? Eliminates the high operational expenditure (opex) of provisioning thousands of ATM PVCs.
? Management tools and staff required as IP/ATM/Sonet/SDH layers must be managed separately.

ATM vs Ethernet
In many cases, idle ATM and Sonet/SDH capacity exists in the network that carriers would like to use, rather than investing in new forms of backhaul. However, there are carriers maintain it is important for them to reduce the concentration factor or the contention in the DSLAM to provide for new services. and this is a strong point for Ethernet (See Table 1).

Table 1: ATM vs. Ethernet

Another concern is that there is substantial process investment in the ATM backhaul related to service and network provisioning that carriers do not want to part with. Opposed to this view are carriers that seek aggressive rollout and growth on the back of the low rollout costs that Ethernet gives.

Similarly, carriers using ATM backhaul have no learning curve for their staff. They know the ATM gear, they are familiar with the equipment, and they have deployed it for a number of years. However, some carriers are prepared to argue that the advantages of Ethernet are so great that its adoption is eventually inevitable and it is better to do it sooner rather than later.

The fourth issue is service related. Proponents of ATM-centric solutions often say that derived voice is a key element of their business plan and that they would like to provide voice over ATM. Carriers more interested in Ethernet-centric solutions often see VOIP as the key element, and they realize that means large amounts of bandwidth for which they need to find a cost-effective solution.

Finally, the evolution of new technologies depends on the vision of different carriers and their long-term network goals. This influences attitudes towards the two technologies. So, a migration with an ATM backhaul approach often has as a target a fiber revolution, ending up with an ATM passive optical network (PON) and ATM VDSL. An Ethernet-backhaul approach is often the starting point for a journey towards an Ethernet-in-the-first-mile solution, with fiber deeper in the network and EFM as a target migration for that fiber revolution.

Ethernet-based distributed DSLAM architecture
As the Ethernet DSLAMs evolve, deployment of Mini-DSLAMs is increasingly popular. A mini-DSLAM allows small, cost-effective equipment to be deployed in remote locations, allowing greater service coverage. This approach is highly expensive with traditional large DSLAMs. Mini-DSLAMs lend themselves well to the distributed architecture.

Next-generation distributed DSLAM architectures use Mini-DSLAMs. These Mini-DSLAMs provide various interfaces, allowing a variety of subscriber-side interfaces to be supported by simply populating Mini-DSLAMs with such appropriate subscriber interface devices as ADSL-2, VDSL or Fast Ethernet transceivers. In addition, a variety of uplink interfaces should be supported, most importantly dual Gigabit Ethernet (GE) uplinks since this is a commonly used configuration.

These DSLAMs are designed to be scalable, from single low density Mini-DSLAMs to large central office DSLAMs that support thousands of lines. Beginning with the universal line card, a mini-DSLAM can be built by simply adding the appropriate mechanical form factor and uplink option to the card.

Mini-DSLAMs are stackable by daisy-chaining them together via their uplink interfaces. In the case of Ethernet uplinks, they can also be aggregated through a separate external Ethernet switch. Finally, the same basic architecture (and perhaps even the exact card design) can be placed into a shelf with a gigabit Ethernet switch fabric to configure a large-scale DSLAM.

Figure 4: Distributed DSLAM Architecture

Advantages of distributed DSLAM
There are advantages of a distributed architecture that have led to its recent popularity. It allows a single card design to be used in both large DSLAM shelves (as a plug-in line card) and small Mini-DSLAM form factors (as a standalone remote DSLAM). The Mini-DSLAM form factor enables a convenient "pay-as-you-grow" approach, where multiple boxes can be "stacked" to add subscribers.

Local traffic processing can be done in the Mini-DSLAMs, which enables many capabilities. It allows configurations in which a line card is remotely connected via Gigabit Ethernet (GE) links, and logically appears to the uplink card as a local line card. Without local traffic processing, flow control at the line card level becomes complicated (must be centralized).

It also allows spatial and logical multicasting to be done at the line card, saving fabric bandwidth and simplifying uplink card multicasting. The uplink card needs to multicast only on a per card, rather than a per-connection basis.

It enables local switching of traffic such as peer-to-peer traffic, saving on backhaul bandwidth use and potentially even backplane use, for cases in which the users are served by the same line card). And it results in an inherently more scalable, "pay as you grow" shelf-based solution due to reduced uplink card requirements.

The entry cost of a DSLAM shelf is much lower in cases where a shelf is used only partially. For example, if only two "dumb" line cards are populated in a 16-card centralized shelf, the cost of redundant, higher cost uplink/fabric cards can represent the majority of the initial system cost. In contrast, deploying two "smart" line cards with lower-cost uplink cards brings substantial relief to this cost structure and allows the initial shelf cost to start low and grow in a more linear fashion as line cards are added.

An architecture with an intelligent traffic-processing element on the Mini-DSLAM can be much more easily adapted to the evolving requirements of the market than centralized architectures with fixed-functions Mini-DSLAM traffic processors. An intelligent traffic processor on the Mini-DSLAM can enable any new feature to be added by simply updating the SW while also avoid expensive field hardware upgrades.

As increased packet processing and throughput requirements are added to the DSLAM, the general market trend is towards the distributed architecture. This architecture is now economically feasible due to low-cost access network processors specifically targeted towards this application.

These factors result in a trend towards shelf architectures based on Ethernet backplanes rather than more expensive switch-fabric based backplanes. Legacy bus-based backplanes are no longer desirable due to the requirement for more throughput on the backplane.

Mini-DSLAM processing requirements
The ideal Mini-DSLAM must possess a number of features to meet demands of next-generation DSLAMs.

Flexibility in configuring both the port interface types and number of ports is vital to offering a scalable architecture upon which any DSLAM can be built. This type of port flexibility ensures compatibility with any type of DSL transceiver interface. Furthermore, it allows the same base line card architecture to be offered across a number of product variants.

Throughput capability should be able to meet the needs of today's applications as well as tomorrow's more demanding, faster applicationswithout requiring major software rewriting or repartitioning.

Processing support
Support for a wide range of processing capabilities, while not adversely impact throughput, for example: Capability to meet the requirements of existing and future applications, from high-speed internet access only to full-featured triple play services which require more sophistication in traffic scheduling, shaping and modification. New requirements that arise should be possible to implement via software upgrade only.

Costthe features mentioned above have to be delivered at a cost point that makes deployment of the service economically viable. The mini-DSLAM must enable a low-cost system solution by incorporating an on-chip control plane processor, using low-cost external memory, and providing a package footprint that allows simple routing, enabling low cost printed circuit board designs.

The above items comprise some of the more common requirements. Based on the requirements presented, it is clear that a number of key processing capabilities are required by the Mini-DSLAMs. Beyond the basics, other capabilities can be present, such as:
? Packet classification with deep packet inspection
? Flexible packet modification and CRC generation
? Per-flow traffic processing
? Comprehensive policing and statistics capability
? Buffer management, congestion control and traffic scheduling capabilities

Given the variety of processing options that a DSLAM line card needs to support, and considering options will continue to evolve as the network matures, a programmable processor is the only way to ensure a future proof architecture. Furthermore, a key enabler to such architecture is a software framework that allows rapid application development while providing enough flexibility to add new features to an application, all while consuming relatively few lines of code.

Future next-generation access networks
DSLAM as an IP routerThis is basically a DSLAM incorporating an IP router. Here subscribers' PVCs are terminated and routed, and all IP traffic is aggregated onto a single Layer 2 uplink, although individual subscriber identities may still be determined at the BRAS by the source IP address (this is BRAS dependent). PPPoE cannot be used across the router to the BRAS. It works well in small networks with no BRAS and a single ISP, or for carriers that simply do not want to provide IP services.

DSLAM as a bridge with VLAN mappingHere the DSLAM bridge applies Ethernet framing and appends a different 802.1q VLAN tag for each subscriber. The Ethernet switches basically tunnel the connection transparently with no knowledge of the individual 802.1q tags, so there is no need for any kind of provisioning in itself, and individual subscribers are not separated by the BRAS at Layer 2. This approach would work well typically with PPPoE and other Layer 2 protocols because of the Layer 2 transparency. It is possible to offer a simple form of QOS for business subscribers by using the 1q tag as an identifier.

DSLAM as a straight bridgeThis is very straightforward, as it is a pure Ethernet model, and no logical subscriber separation is maintained at Layer 2, so Layer 2 is bridged transparently through the network. PPPoE and other Layer 2 protocols may be used as well. This architecture works very well with PPPoE, and the subscriber identities are maintained by PPPoE session IDs.

DSLAM as an L2TP LACIn this model, DSLAMs terminate PPPoE tunnel sessions inside L2TP tunnels. So the session for all ISPs uses the same tunnel, and carriers can obviously create backup tunnels for resilience. Centralized BRAS farms terminate tunnels from DSLAMs. The entire network between the DSLAM and the BRAS is IP based.

It is evident that with more flexibility, scalability, low equipment cost and increased bandwidth, Ethernet-based access multiplexers are gaining popularity. Also with the growing complexity and increased packet processing and throughput requirements, there is an increasing need of a cost effective distributed DSLAM architecture in which much of the traffic processing requirements are performed in the first mile. This can be achieved through distributed DSLAM network architecture, a key component of this architecture being Mini-DSLAMs.

These Mini-DSLAMs should have high processing and throughput requirements, that make them essentially very close to the traditional centralized DSLAMs. They should have the capability to meet the requirements of existing and future applications, from high-speed Internet access to full-featured "triple play" services.

About the author
Sachin Kumar Jain
is an STMicroelectronics engineer located in Noida, India. Sachin can be reached at

Article Comments - Overview: Distributed IP/Ethernet DS...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top