[May 14, 2001]

You Can Have Your ATM-Based VoDSL QoS, And IP Too

BY BRYAN LONG AND JOHN REISTER

VoDSL. In the acronym-happy world of telecommunications, these five letters -- which stand for Voice over Digital Subscriber Line -- spell opportunity for broadband providers.

Cahners In-Stat Group projects that U.S. VoDSL service revenues will approach $4.2 billion in 2001 and $9 billion by 2003. Small- and medium-size businesses (SMBs), already attracted to DSL by the favorable economics of high-speed data services over copper-pair phone lines, are ideal prospects for VoDSL service. According to Cahners In-Stat, there are currently more than 6.2 million SMBs in the U.S. Cahners In-Stat also forecasts that total SMB telecom expenditures will exceed $66 billion in 2001 and reach $81 billion by 2003.

As broadband providers prepare to capture part of this spending by offering cost-competitive VoDSL service to SMB telecom customers, another industry acronym rears its formidable head: QoS (Quality of Service). You can't do VoDSL without it.

In today's telecommunications environment, at least a couple of truths seem self-evident: Asynchronous Transfer Mode (ATM) technology still dominates the network's core, and ATM is still the prevalent QoS enforcement mechanism for broadband voice traffic. This has led some observers to infer that ATM's current ascendancy at the network's core compels ATM-based DSL access solutions at the network's edge, and that only ATM-based access solutions can support QoS for broadband voice.

These inferences may have been valid a few years ago, but they aren't today. While ATM has continued to dominate the network's core, a revolution has erupted at the network's edge, where a rising tide of data traffic has fueled the ascendancy of Internet Protocol (IP). Today, even voice traffic is increasingly treated as data, transportable in IP packets. Providers with the right access solutions can leverage IP to achieve considerable gains in provisioning cost-efficiency for all services -- including voice -- and tremendous gains in network scalability. The right access solutions for today's packet-dominated networks are DSL concentrators with IP service intelligence.

ATM's PVC Problem
IP packets' headers are rich with information that identifies the precise nature of all broadband traffic, its source, and its destination. ATM-based DSL Access Multiplexers (DSLAMs) cannot see these headers, because they slice IP packets into 53-byte ATM cells. Operationally, this blind spot forces a DSL broadband provider with ATM-based DSLAMs to provision separate, end-to-end Permanent Virtual Circuits (PVCs) for every subscriber and for each kind of traffic for every subscriber.

In contrast, IP-optimized DSL concentrators see whole Internet packets, including their headers, and differentiate one kind of packet from another. They can make intelligent traffic-forwarding and queuing decisions in the access network based on IP addresses or applications, prioritize traffic based on IP header or application information, and aggregate like kinds of traffic from different subscribers onto shared virtual circuits through the network backbone.

IP-optimized DSL concentrators' subscriber-aggregation capabilities yield significant provisioning and scalability dividends. Consider, for example, a provider with 100 customers, each subscribing to three broadband services. If the provider serves these customers with an ATM DSLAM, he must provision and maintain 300 PVCs -- one per service, per customer. But if the same provider deploys an IP-optimized DSL concentrator, he need only provision and maintain three PVCs.

The provisioning and scalability advantages of IP-optimized solutions over ATM-based solutions in the access network appear decisive. Nevertheless, some still argue that when it comes to voice no access solution does QoS like ATM. Certainly, no other solution does it as clumsily. ATM-based DSLAMs assure QoS for broadband voice traffic by assigning each call to its own PVC and designating QoS on a per-circuit/per-call/per-subscriber basis -- further compounding ATM's inherent PVC proliferation problem.

DSLAMs Do QoS Better
DSL concentrators with IP service intelligence not only do voice QoS as well as ATM DSLAMs, they do it better, because they don't inundate the provider's network with PVCs. Working with compatible Integrated Access Devices (IADs), IP-optimized DSL concentrators deliver toll-quality voice service through a multi-dimensional approach to QoS that includes:

• DSL Priority Queuing: Real-time and non-real-time class-based queuing puts voice traffic in the high-priority queue in the IAD and the DSL concentrator line card, while placing data traffic in a low-priority queue. Voice traffic is serviced first, and data is transmitted only when there is no voice queued up. Voice traffic is identified either by having the IP Delay Bit set, or by virtue of the PVC it is to be sent on. DSL priority queuing assures that voice traffic gets the priority it needs.
• Frame Fragmentation: Frame fragmentation enforces QoS when a voice packet arrives at the voice queue just after a large data packet begins to leave the data queue. When this occurs, the data traffic is fragmented on the fly, and the voice traffic is allowed to proceed. Frame fragmentation keeps the average DSL transit delay for voice traffic as low as 5 ms.
• ATM QoS on the WAN: Aggregated voice traffic from many subscribers rides over the WAN ATM link within a variable bit rate (VBRrt) circuit, while aggregated data is transported on a unspecified bit rate (UBR) circuit. Class-based queuing on the WAN ATM link ensures uncompromised voice service quality.
• Packet Level Discard: DSL concentrators with IP service intelligence utilize packet-level discard, a highly efficient mechanism to avoid transmitting fragments of a packet over an ATM network, which will only be dropped at the receiving end anyway.
• Admission Control: IP-optimized DSL concentrators implement an admission control feature that allows the service provider to avoid oversubscribing the demanding needs of voice services.
• Congestion Management: In the unlikely event that the concentrator is overloaded with packets, low-priority packets are dropped first.

This multi-faceted approach to quality of service enables an IP-optimized DSL concentrator to maintain voice QoS while aggregating voice traffic from many subscribers' IADs onto a single virtual circuit to the network backbone. Where the provider with an ATM DSLAM must set up 100 end-to-end PVCs to provision 100 voice calls for 100 subscribers, the provider with an IP-optimized concentrator can provision the same number of calls with only one aggregated virtual circuit.

Comparing Costs
The considerable difference in the number of circuits required to provision multiple services over ATM-based DSLAMs versus IP-optimized concentrators translates into big differences in upstream capital and operations costs.

An ATM switch costs between $150,000 and $250,000, depending on configuration. A typical ATM switch can handle 12,000-15,000 best-effort, UBR (Unspecified Bit Rate) virtual circuits. A provider who relies on ATM-based DSLAMs to offer best-effort Internet access to 20,000 subscribers will need to invest in two ATM switches, at a cost of $300,000 to $500,000. If the ATM-based provider augments best-effort Internet service with a VPN, he needs to buy two more ATM switches to handle an additional 20,000 PVCs, swelling his initial capital costs to somewhere between $600,000 and $1 million.

Now suppose the provider adds VoDSL services for 20,000 subscribers using GR-303 packet voice technology. The provider will need to provision another 20,000 PVCs. These virtual circuits need to be rate-shaped with controlled QoS and provisioned as VBR-rt (Variable Bit Rate - Real-Time) circuits. ATM switches' capacity drops substantially when the circuits have a guaranteed bit rate. Instead of 12,000 to 16,000 PVCs, they can only handle 4,000 to 8,000 PVCs. The provider now needs to buy three more ATM switches to provision VoDSL service, boosting his up-front costs to between $1 million and $1.7 million.

IP solutions reduce the provider's capital costs by a factor of almost seven. Suppose the provider deploys 125 192-port IP-optimized DSL concentrators to provision best-effort Internet access and VPN service for 20,000 subscribers. The concentrators' subscriber-aggregation capabilities enable the provider to provision these services with 250 virtual circuits -- two from each concentrator -- across a single ATM switch. When the provider adds IP-based VoDSL service, he only has to provision one additional VBR-rt circuit from each concentrator. The provider now has 125 VBR-rt virtual circuits and 250 UBR data virtual circuits, all of which can still be handled by a single ATM switch. The IP-optimized provider's total up-front investment is $150,000-$250,000 -- a savings of $850,000 to $1.5 million.

The IP-optimized provider also reaps ongoing cost savings. With fewer ATM switches, the provider needs fewer pipes to backhaul voice and data traffic. And with DS3 backbones running $2,000 a month and OC3s $5,000, the provider can save tens of thousands of dollars in monthly backhaul costs.

Further Arguments For IP
IP-optimized providers enjoy another important advantage: Their network-access solutions are essentially future-proof. Network cores are changing. ATM technology has lost the most important advantage it had in core networking: speed. ATM used to win over IP routers because only ATM could scale to speeds above 45 Mbps. Now, high-performance, silicon-based IP routers enable dynamically scalable speeds from 1 Megabyte per second to 1 Terabyte per second and are proliferating in metropolitan area and wide area networks. These next-generation technologies can move both toll-quality voice and data at a small fraction of the cost of ATM. They use policy-based management technologies like DiffServ and Multi-Protocol Label Switching (MPLS) to enforce QoS and support Service-Level Agreements (SLAs) for all services, including voice. Eventually, all traffic -- voice and data -- will move across these networks in IP packets. ATM-based DSLAMs will not migrate to this new networking environment; IP-optimized concentrators will.

Finally, it is clear that IP will dominate the next generation of user devices and services: LAN-based and wireless telephones with integrated Internet browsing, PCs with high-quality videoconferencing support, multicast music and video services, and Web-based communication portals that allow you to program custom calling features. Innovative and profit-generating new IP services will require an end-to-end IP architecture with IP QoS.

In other words, you can have your VoDSL QoS cake and eat your IP cake too, even in today's ATM-dominated networking environment. And you can migrate seamlessly to next-generation all-IP networks, smiling all the way to that other ATM at the bank.

Bryan Long is vice president of marketing and John Reister is vice president of strategy and advanced technology for Copper Mountain Networks, Inc. Copper Mountain Networks manufactures DSL equipment for central office, digital loop and multi-tenant unit (MTU) broadband networks worldwide. Its DSL solutions enable carriers and service providers to deliver cost-effective, high-performance data and voice services over existing copper telephone wiring.