April 2000

BY STEPHEN DUFFY

Any farsighted enterprise depending on the Internet for success wants a business-class, multi-service IP network optimized for quality voice, video, and data � a network with bandwidth quality of service (QoS) comparable to the PSTN. Such next-generation packet networks need platforms that can grow quickly and elegantly to meet exploding bandwidth demand. They also must incorporate value-added, differentiated services such as high-bandwidth allocation and high-speed transmission � and support QoS at gigabit and terabit rates.

While QoS is used sparingly today in emerging IP networks, farsighted network builders should recognize that surging appetite for beyond-best-effort applications, such as VoIP service, soon will oblige prevalent QoS for competitive survival. Designed to be the core of these new nets, next-generation network elements (NEs), such as terabit switch routers, will be relied upon to not only assure adequate bandwidth but also enable QoS that can minimize management costs and add revenue-producing service offerings.

Reliable Internet QoS is essential to developing service level agreement (SLA) revenue models in which a business can buy only the type of service it needs � with billing commensurate with use. A business-centric Internet should provide QoS levels to match different classes � akin to airlines offering coach/business/first-class service options that will accommodate the range of typical business traffic, from low-priority Web surfing to mission-critical voice transmissions. And as new applications like real-time packetized voice and audio/video streaming migrate to the Internet, the network�s operational requirements must change to accommodate them.

Such traffic variety will challenge resources and create revenue-enhancing opportunities. Delay-tolerant, highly elastic applications like e-mail and file transfer protocol (FTP) are more tolerant of network delays than real-time, delay-sensitive interactive applications such as VoIP, which demand low loss and bounded delay and jitter. Real-time applications must receive data within some finite period of time or the packets essentially become worthless. This makes it mandatory to enforce an upper bound of delay for such flows. Users should expect to pay a premium for such guaranteed precision. Conversely, a predictive service may deliver a fairly reliable but not guaranteed level of service at a much lower cost.

Note that the emerging demands of next-gen networks do not necessarily revolve around bandwidth. The strategy of deploying an increasing number of higher data-rate channels over dense wave division multiplexing (DWDM) fiber has not and will not quench the insatiable demand for bandwidth. In fact, it inevitably will lead to a commoditization in the cost of bandwidth, so the capital-intensive solution of overbuilding networks won�t suit a world where the cost of bandwidth is dropping faster than network transmission costs.

Over-provisioning can relieve congestion, but it can�t deliver tightly bounded delay mechanisms for guaranteed SLAs that incorporate delay-intolerant applications such as corporate data centers or VPNs. IP networking in which carriers can offer �always on� types of services, readily differentiate and prioritize traffic, and efficiently manage their networks while minimizing infrastructure costs, will need �intelligent� next-gen NEs that can flex new flow-based protocols and QoS mechanisms.

NEW METHODS
Differentiated services (DiffServ) and Multi-Protocol Label Switching (MPLS) standards, developed by the networking industry�s Internet Engineering Task Force (IETF,), are linchpins to providing QoS over the Internet. They hand service providers the appropriate means to control high-priority applications, like VoIP, and guarantee quality end-to-end service.

DiffServ is an IP QoS mechanism that allows network traffic to flow with sufficient quality to ensure successful deployment of the application. It enables identification of the traffic class of each packet, the primary component allowing QoS to be deployed, so traffic can be directed without the overhead of managing each end-to-end flow. DiffServ is based on common, configurable parameters derived from the application�s bandwidth, delay-jitter, and loss-probability requirements.

QoS on the Internet obliges consolidating traffic flows, which must be summarized into aggregate groups or classes across the network (much like routing protocols provide network summaries to the Internet�s routing table). DiffServ aggregates individual flows and marks packets for the appropriate level of service available from an assortment of predefined packet treatments.

Because it affords nearly boundless combinations of services, a differentiated-services network can easily provide the service levels demanded by either new-age enterprises transacting business entirely over the Internet, or traditional businesses migrating from private voice/data networks to IP-based service provider nets.

Meanwhile, network designers must be sure the amount of high-priority traffic put on any individual network link does not exceed its bandwidth. MPLS virtual connections among packet switches � Label Switched Paths (LSPs) �can be routed independently of the underlying link topology. This dynamically computes routes for high-priority traffic flows based on a predetermined set of network policies. Controlling traffic flows based on such network constraints means all links share traffic load efficiently and equally with none becoming over-committed with high-priority traffic.

The MPLS standard is based on the blending of connectionless protocols, like IP, with virtual-circuit networking concepts. Compared to using ATM, MPLS is a more efficient way to engineer traffic. Without it, there is no way to ensure high-priority traffic won�t exceed the physical capacity of any given link in the network. By using MPLS to define the physical links that specific traffic will use, service providers can manage traffic loads to ensure they are within any link�s ability to handle with acceptable quality.

NEW MODELS
Internet QoS is achieved through three main processes: Queue management within a single network element, performing things like shaping, scheduling, weighted fair queuing, and random early detection; controlling flow end-to-end to ensure that high-priority traffic avoids congested nodes; and marking flows to signal different treatment, then aggregating multiple flows into classes across the backbone. Next-generation network elements need all of these features, which terabit switch routers deliver.

Terabit switch routers already provide massive hardware-forwarding capabilities and intelligent scaling capacities that can maintain internal protocols as the network grows. Their repertoire covers QoS mechanisms that include sophisticated tools for network-congestion management and intelligent-buffering or congestion-avoidance. They perform admission control, flow identification/classification, traffic policing, and scheduling to avoid network congestion and ensure that traffic flows conform to their pre-negotiated QoS contract. And router vendors, naturally, continue to work on creating technology for next-generation IP networks that will take Internet QoS to still higher levels.

Service providers can use intelligent terabit switch routers to leverage QoS and MPLS to both gain operational network efficiencies and enable higher value-added pricing models that are founded on network-aware treatment of applications. Differentiating traffic by user and application means network resources can support more granular levels of service. New service levels, meantime, will give network operators the flexibility to offer different pricing models, i.e., a pay-as-you-go or X+Y (flat rate for a fixed number of hours, variable usage thereafter). The end result is more ways for network operators to generate incremental sources of revenue.

NET TALK
VoIP will both benefit from and rely on the QoS capabilities of the next-gen routers which will make up the new packet-network infrastructure. Scalable, flexible devices will allow emerging facilities-based carriers to offer innovative services in order to compete against larger, more established players.

In fact, �next-gen� is fast becoming �now-gen� as carriers already are rolling out or taking to trial next-generation networks which have IP running on top of optical cores (obviating the need for SONET equipment in between). The trend already underway in switching � for both the LAN and WAN markets � is the advent of multi-layer capabilities able to provide QoS at wire rate to diverse network flows. Soon to come will be new standards to increase interoperation of the IP service layer and the underlying optical layer. The auto-provisioning in these next-gen networks will need switch hardware capable of detecting network bottlenecks and increasing service capabilities in real time, able to dynamically add or redirect bandwidth as network loads change.

The combination of flexible, hardware-based, QoS-aware switches, DiffServ for marketing and aggregating flows, and MPLS for traffic engineering will be bringing sophisticated routing capabilities to packet networks. This means such connectionless entities can behave in a deterministic fashion. And when a network performs in a predictable and controlled manner, applications of any priority can be supported on a common infrastructure.

By bringing such intelligence to the Internet, terabit switch routers enable the success of high-priority applications such as VoIP. They provide an abundance of new traffic-engineering mechanisms and QoS capabilities that enable policy management and application-aware traffic forwarding. The net result is successful delivery of high-quality services that efficiently utilize network resources, to the benefit of service providers and end users alike.

Stephen Duffy is a product line manager with Avici Systems, Inc., of Billerica, Mass., and can be reached at [email protected]. With its mission to build �speed of light� networks for the 21st century, Avici is the leader in integrating packet-based technology with carriers� optical investments to ensure highly scalable, highly reliable, and highly cost effective networks for the future. For more information, visit the company�s Web site at www.avici.com.

BY GARY ROGERS

Packet telephony has been lauded as the technology that will enable service providers to enhance their competitive position through reduction of operating costs and the creation of new revenue streams. Early on, the market clearly recognized a significant opportunity for technologies that would enable the convergence of the voice and data worlds, and the industry has witnessed the evolution in packet telephony solutions in a relatively short period of time.

Originally, basic connectivity was provided via Sun or Windows NT servers equipped with Dialogic or NMS VoIP cards. These solutions helped to advance the market while delivering �proof of concept� that packet telephony would actually work. However, early products such as these were typically not well suited for carrier-class deployment, as they lacked the reliability and scalability required in the service provider market, and were also rather expensive on a per port basis.

Remote access vendors also targeted this opportunity by adding VoIP capability to their existing remote access server (RAS) platforms, which was considered a logical extension, since the RAS box was already handling Internet calls. However, these platforms were designed with modem traffic applications in mind � they fell short on the capacity front, typically supporting fewer than 300 ports. Additionally, they generally did not support System Signaling 7 (SS7), as they were not designed for VoIP applications from inception. The RAS with VoIP may have been suitable for small ISPs, but was certainly not robust enough to address carrier-class requirements, particularly when an entirely new voice network was being built out.

Clearly, in the migration to the converged voice/data network, what was needed was a carrier-class solution that bridged the gap between both worlds. Within the past year, gateway switches have entered the market and the response to these new solutions has been extremely positive. Nearly all carriers are planning for packet voice, a significant number have products in trials, and a handful of leading-edge service providers have actually selected platforms and begun deployment of their next-gen voice infrastructures.

One might wonder why this new breed of product has met with early success, while previous solutions were not accepted into carrier environments. A major factor is these gateway switches have been designed to address the stringent requirements of service providers, delivering:

• Absolute reliability: 99.999 percent uptime;
• Scalability to support large call volumes;
• True toll-quality voice transmission;
• Full interoperability with the PSTN, including support for SS7;
• High-density, compact footprint that conserves expensive Central Office space;
• Openness to accommodate a wide range of enhanced service platforms; and
• Flexibility to support current and future caller services (both PSTN and IP-based).

DELIVERING THE COMPETITIVE EDGE THROUGH SERVICES
To take a leadership position, a carrier must offer services people are willing to pay for, and get those services to market quickly. Packet telephony helps. First, it can facilitate applications that are difficult or impossible to deploy when voice and data ride separate networks. Second, it can actually enhance time-to-market for new applications.

Certainly, carriers will need to offer the services already common in the voice network: Voice mail, call forwarding, caller ID, etc. But the explosive growth of the Internet also demands services that will bridge the two worlds. Examples of such services include unified communications, Internet �click-to-talk,� and more. These have been slow to develop, in part because of the separation of networks. It is widely held that network convergence will eliminate the barriers and yield a flood of innovative new services, many of which have not yet been imagined.

Packet telephony enables carriers to employ �best-in-class� services, including:

Existing SS7 services. Many carriers have services in their SS7 SCPs. Softswitch support for such applications allows their immediate use on the packet telephony network and allows use of common databases between the circuit switching and packet telephony networks.

Carrier-developed applications. Carriers may wish to develop certain services, either because the application does not already exist in the market or because the carrier has special expertise in the subject. Open APIs for routing and service provisioning applications, as well as application server applications, allow the carrier to influence every aspect of network operation.

Services based on the gateway switch capabilities. Some services can be implemented using only the functions provided by the gateway switch, such as tone generation, digit collection, and announcements. In these instances, creating a new application is as simple as creating the softswitch script and provisioning the service for one or more users.

Third-party applications. Use of open APIs exposes the carrier market to a wide range of third-party application providers. The service possibilities are unlimited, ranging from sophisticated PSTN applications such as prepaid calling cards to IP or converged network applications such as �click-to-talk.�

Little more than a year ago, these types of service options were completely unheard of. The combination of the carrier-class packet telephony solutions that have emerged in the past twelve months and the Internet model of service delivery will have a long-lasting impact on the telecom market. Not only will it help carriers improve their competitive position, but also will provide users access to applications that will significantly improve their business and personal lives.

Gary Rogers is vice president of sales and marketing for Sonus Networks. Sonus was founded in 1997 with the mission of building carrier-class network solutions to facilitate the movement of telephony from circuit networks to packet networks, and to enable a new generation of packet-based applications for integrated voice and data services. Sonus Networks� products fully interoperate with and extend the life and utility of today�s public network. For more information, visit the Sonus Web site at www.sonusnetworks.com.