The Internet and Internet Protocol (IP) networks have grown tremendously during the past decade; fueled primarily by best-effort applications such as e-mail, Web browsing and noncritical data services. Despite this impressive growth, IP networks have not succeeded in supporting real-time applications and converged legacy services, which demand greater levels of stability and reliability. As service providers look to decrease the number of backbone networks, they require a routing solution that supports the graceful migration of legacy Layer 2 services to IP while providing the stability, quality-of-service (QoS), reliability and convergence improvements to support new real-time IP services.

Finances and technology are driving the migration to an IP-based network. New IP services are critical to improving top-line revenue. Decreasing the number of networks needed to support legacy and new real-time services is essential to improving profitability. And growth through IP, rather than maintenance of the legacy system, is laying claim to capital expenditure dollars these days. From a technology standpoint, new IP/multiprotocol label switching (MPLS) routers are capable of delivering telco-grade availability, ATM-like QoS, convergence improvements and scale. Debate no longer centers on whether IP is the de facto convergence platform. Instead, the question has become, How long will it take, and which vendors will take us there?

Service providers recognize that profitability depends on achieving higher margins from real-time services and cost efficiencies of converged networks. Providers need to upgrade their IP networks to support not only existing best-effort services but also real-time services and existing Layer 2 services. Best-effort services such as e-mail are able to withstand the significant delay, packet reordering and outages common on most IP networks. But real-time services, such as voice-over-IP (VoIP), video, streaming media and interactive gaming, demand a higher level of network performance with low latency and high network availability. Essential services that currently reside on ATM and frame-relay switches cannot be transitioned to a network that neither is stable nor delivers the QoS service-level agreements require.

To understand the obstacles that prevent support of real-time services such as VoIP, we must remember the original design goals of the Internet. Arpanet, the precursor to the Internet, was developed with the primary goal of survivability. Fundamental to that goal is the ability of IP networks to reroute in response to network changes. The design makes IP networks highly resilient to outages but also subjects them to frequent disruptions and changes in routing.

Those network disruptions were acceptable for popular best-effort applications such as e-mail and Web browsing because those applications exist independently of variations in latency and throughput. But such new applications as interactive gaming, VoIP, music and video-on-demand are already commercial services for carriers. Industry reports say that VoIP, gaming and streaming applications already account for more than 10 percent of the IP traffic in the United States. The newer class of services will not tolerate the service interruptions that are accepted in best-effort traffic. As real-time applications continue to increase, network outages will become more visible, and service providers will have to react or lose customers to other carriers.

For IP networks to support demanding real-time applications and converged legacy networks, carriers must overcome five obstacles: poor router reliability, lack of link protection, disruptive operations, slow convergence time and multiservice support.

Router reliability

Two primary causes of IP network downtime are router hardware and software failure. In contrast to traditional central-office equipment such as voice and ATM switches, IP routers were not designed to support carrier-grade 99.999 percent availability. Typical large IP networks achieve between 99.95 and 99.99 percent availability. This is 10 to 50 times more downtime than the "five nines" availability benchmark of legacy data networks.

To accommodate the lack of telco-grade reliability, carriers have become accustomed to the costly practice of duplicating core routers in their points of presence to compensate for the lack of reliability in a single router. Next-generation routing platforms have achieved 99.999 percent availability by providing full hardware and software redundancy in single router. A key component is delivering fail-safe route controller protection. Nonstop routing implementations use one or more backup route controllers to maintain the pertinent routing state and supporting adjacencies with surrounding routers, preserving connectivity to peers. Because this approach is self-contained, it requires no interoperability with other routers in the network. In-service software upgrades, whereby a carrier loads a new software image on a backup route controller while the primary continues to operate, are also a requirement. Hitless software upgrades are essential to the delivery of 99.999 percent availability because they eliminate router downtime associated with software upgrades. Where global enterprises are concerned, maintenance windows based on time zones are no longer acceptable.

Link protection

A study by the University of Michigan observed that link failures accounted for 32 percent of the outages in a large regional IP network. That's not particularly surprising, since IP networks traditionally have been built without locally protected links. Instead, these networks have relied on the ability of routers to route around failed links. But there's a glitch: The approach produces unacceptable disruption to real-time and converged services.

To avoid triggering protocol convergence and affecting services, link protection must occur locally and achieve failover in less than 45 milliseconds. Many local link protection schemes, such as MPLS Fast Reroute, link aggregation and Sonet APS, are now available. Powerful new Sonnet/SDH link aggregation mechanisms such as composite links enable up to 64 physical links to be grouped into a single logical link. In the event any member link fails, traffic is redistributed across the surviving links in less than 45 ms.

Another major reason IP networks fail has to do with operational changes, such as software and hardware upgrades, link expansion and configuration changes. It is not possible to perform those routine tasks on traditional routing platforms without causing disruptions in service.

A new generation of core routers offers in-service hardware scalability and hitless in-service software upgrades. By migrating to such platforms, carriers are able to eliminate hardware and software upgrades as a source of disruption in the network. For example, nonstop routing technology allows carriers to perform in-service software upgrades and revert to previous stable configurations without interrupting services.

Protocol convergence

By making changes in the areas outlined above, carriers could dramatically reduce the number and frequency of disruptions to an IP network. Even with such precautions, however, no network is completely immune from human error or topology changes.

In large IP networks, service interruptions that last less than a minute are commonplace and not typically calculated in availability reports. Because end-user applications such as e-mail or Web browsing forgive such outages, designers have paid less attention to those brief service interruptions than to other areas. But for mission-critical applications that are sensitive to microbursts, even a short interruption presents serious consequences: A VoIP call will be disconnected or an interactive gaming session disrupted.

It is hard to put a dollar value on these fluctuations, but the overall impact on the user's experience and dissatisfaction with the service provider's service is measurable. If interruptions occur frequently, users will find another provider that can deliver the stability they need.

When failures or changes do occur, the network must minimize the impact by rapidly converging IP protocols. Protocol convergence involves recalculating the shortest paths to all destinations, and then updating the routing tables used to forward packets. The process, called convergence time, can range from 10 seconds to several minutes, depending on the router, network size and protocols involved.

When selecting a routing platform for the real-time IP network, a critical consideration is convergence time performance, which needs to occur in less than 5 seconds. The strategies used to reduce protocol convergence times may include the optimization of convergence algorithms, configurable hold-down timers for Sonet and IGPs, and simplified LDP interaction and elimination of redundant lookups.

Quality-of-service

Another focus area for real-time services and convergence support is implementing router QoS in hardware. Real-time services require stringent latency and jitter for all protocols and advanced features such as filtering, sampling and ACLs, without performance degradation.

If a carrier is unable to use a feature because of performance degradation, then it is not useful. Another requirement is keeping packets in order because real-time applications such as VoIP and video-over-IP, including MPEG-2 and MPEG-4, are sensitive to reorder delays. A reorder delay produces an underflow of packets, which may appear to the user as a freeze frame for video or a dropped connection for voice.

It is unwise to rely on the best-case numbers from lab testing that some router vendors will use to illustrate how their service performs. Always measure latency and jitter in an operational environment; those numbers can be substantially higher than in isolated lab tests. QoS must also support IP Unicast, IP Multicast and MPLS.

Once you've stabilized IP routers, the final step is to converge legacy IP services to the high-growth IP/MPLS network. Providing support for legacy traffic on IP routers is the only way that carriers can gracefully migrate Layer 2 services to IP/MPLS. Terminating legacy traffic or encapsulating it are two options for achieving this. Increasingly, routers are using a new generation of network processors to deliver multiservice capabilities to IP routers. Moreover, because they leverage a programmable network processor, they provide any-service, any-port capabilities not available on ASIC or FPGA-based line card designs.

Carriers are reducing the number of networks by converging legacy services to IP and growing top-line revenue through new real-time services. The real-time IP network enables carriers to realize the efficiencies of converging traffic from multiple services over a common backbone.

Esmeralda Swartz is director of strategic marketing at Avici Systems (North Billerica, Mass.). She can contacted at eswartz@avici.com.

See related chart