NETWORK ARCHITECT

Next-Gen Router Prospects

from the May 2002 issue of Business Communications Review, pp. 16–17

Routers, the workhorses of IP networks, have come a long way during the last 20 years. They sell as standalone products, and are embedded in other types of devices, such as ATM switches or firewalls.

At the very low end, routers are commodities; there's significant competition and very low prices for SOHO routers from the likes of Linksys and D-link. At the upper end, Cisco and Juniper dominate the market for Internet backbone routers. And, in between, Cisco dominates; knowledge of Cisco's IOS and router configuration is almost universal, giving the company a virtual monopoly in large and mid-size companies.

The combination of Cisco's dominance and the lack of much new functionality creates a tendency to assume that nothing much is happening with routers. But big changes are on the way, particularly in service-provider networks.

Market Grows And Segments

The bursting of the networking investment "bubble" has had a devastating impact on the equipment vendors. Many have gone bankrupt, and even the survivors are experiencing major difficulties. The providers have significantly reduced capital expenses (CapEx), and are now turning their attention to minimizing operational expenses (OpEx). As a result, the creation of new services—and associated new product features—has to wait until conditions improve.

Most of the capacity "glut" in provider networks has been in optical fiber. As a rule, providers don't "light" fiber until demand requires it; the cost of technology like optical amplifiers, repeaters and DWDM is simply too high. To turn optical channels into usable IP network capacity also requires new routers, which the providers also won't invest in until demand materializes.

Meanwhile, IP traffic continues to grow. To be sure, it's "only" growing at about 100 percent per year, not the 300–400 percent of a few years ago. But still, doubling traffic every year means there'll be long-term demand for more routers, especially those that offer smooth scalability.

Another major trend is for providers to converge their separate overlay ATM/frame relay and IP-routed networks into a single backbone, probably based on MPLS (see this column, March 2002, pp. 18–20). This movement toward convergence is more oriented toward reducing OpEx.

These converged networks will have to support multiple Layer 3 and 2 VPNs and services at the edge, which places routers on a collision course with ATM/frame-relay switches. IP-routing protocols (e.g., OSPF, IS-IS, BGP) will continue to be used, but much (if not most) traffic forwarded by routers will be via MPLS. We may need to think of service-provider routers primarily as MPLS devices rather than just IP devices.

Routers also are becoming more specialized. In the past, service providers (or enterprises) that outgrew the capacity of their core routers could just redeploy these systems at the edge, where traffic rates were lower. Many Cisco 7500s were repurposed as edge routers in this fashion.

But now, edge routers are being asked to do far more than just forward packets; they may be responsible for accounting, MPLS tunnel termination, traffic classification and marking, prioritization, etc.—all of which require packet processing. In contrast, core routers must be "lean and mean," to achieve ever-higher packet forwarding rates; they have little time to do packet processing.

The result is that separate markets have emerged for each type of router. Unisphere has already gained market share at the edge, and startups such as Laurel and TiMetra are specializing in this area. By contrast, newcomers Pluris, Caspian and HyperChip focus on the core.

VPN demand also may create an opportunity to develop routers tailored to the needs of providers who deploy scalable "virtual routers." CoSine and Crescent offer products that can run hundreds or thousands of router instances in software. And for providers that prefer to deploy multiple routing instances in hardware, with separate control plane and forwarding processors per customer or network, Allegro has created the industry's first multirouter in a single chassis.

The bottom line? The router market has fractured into multiple, specialized markets.

Router Issues

But while opportunities may be expanding, router vendors face significant challenges. As noted, providers are more hesitant than ever to spend, and longer-term, it has yet to be proven that they can profitably sell just IP services.

As a result, pressure will build on vendors to sell routers that are multiservice—transport Ethernet, ATM, frame relay or other protocols. But before the vendors can convince carriers they can handle multiservice traffic, they must address a perennial complaint: Routers are not "carrier class." IP router networks do not provide five-nines availability, for several reasons: Unreliable software, the need to take routers out of service for new software loads and the 30 seconds or more it takes for routed IP networks to converge to new routes where there's a link or node failure.

The lack of scalability also needs solving. A typical ISP POP has a two-level hierarchy: multiple edge routers with customer—facing interfaces that are each cross-connected to at least two core routers (for redundancy) that terminate backbone trunks. As customers are added, more edge routers are required, which, eventually, either uses up all the ports on the core routers or causes them to run out of forwarding capacity.

One solution is to add more core routers (or a separate mid-level distribution layer of routers), but this greatly increases cost and complexity. As customers and/or traffic levels increase, the number of unproductive ports (i.e., those that don't interface directly to customers and generate revenue) as a percentage of total ports in the POP increases unacceptably. Router clusters also waste forwarding capacity, as packets undergo unnecessary hops through additional routers.

The only real solution is to buy a bigger router, but providers are tired of having to repeatedly upgrade. The ISPs have had to replace their core routers every 2–3 years to meet traffic growth; in contrast, other carrier voice and data switches last seven or more years. Router products need to be more scalable and/or upgradeable.

Technology Trends

To address these and other issues, router vendors are incorporating new technologies and design enhancements. The improvements include:

• Performance and Scalability: More and higher-speed optical interfaces continue to be added to core routers. These include 10-Gbps Ethernet and OC-192; 40-Gbps OC-768 is coming soon. Router forwarding performance must also improve. The race between Cisco, Juniper, Avici and the startups continues to determine who can deliver the biggest BFR (big "fast" router).
• Hitless Software Upgrades and Fault-Tolerant Operation: To address five-nines availability, vendors are working on new architectures that permit fast failover when a route processor fails. The challenge is not just to continue forwarding packets while re-convergence by a backup processor takes place, but also to avoid the need for re-convergence in the first place. This will potentially reduce the amount of redundancy required in POPs, improving both CapEx and OpEx.
• Improved Convergence Times: Vendors are working to improve convergence times when a link fails; the goal is to achieve SONET—like 50 millisecond or less recovery times.
• Network Edge Processing: With more powerful network processors and classifiers, edge routers will be able to handle VPNs, BGP/MPLS, virtual routers, QOS and accounting.
• Route Scalability: With the increased popularity of multihoming (see BCR, September 2001, pp. 20–22) and continued Internet growth, Internet routing tables continue to grow. Routers must have sufficient route processing capacity and memory to handle future table sizes.
• IPv6: IPv6 has a low profile in the U.S., butdevelopments in Japan and China, and increases in IP-addressable devices (cellphones, game players, etc.) are improving prospects for IPv6. At a minimum, IPv6 may become a "check-list" item for new purchases.

Conclusion

Despite the current market malaise, prospects for router vendors, relative to other network equipment suppliers, are bright. Many service-provider ATM/frame-relay networks are getting tight on capacity, and as the providers consider migration to MPLS-based infrastructures, it creates a new market opportunity for router vendors.

Furthermore, the continued exponential growth of ISP traffic, combined with providers' need to reduce OpEx and the complexity of their POPs, may make new higher-capacity routers attractive—even to providers trying to curb their expenditures. Next-generation routers that can scale, offer non-stop availability and that meet the needs of budget-constrained service providers should find more than enough customers.

Top of Page