Is IP switch routing the last frontier?
BY MEGHAN FULLER
Both optical transport and access networks have achieved exponential increases in capacity and performance-thanks to innovation in WDM and optical switches on the optical side and DSL and cable on the access side. The IP core network, however, is "almost like the final frontier," claims Michelle McLean, director of strategic marketing at Pluris Inc. (Cupertino, CA). It is the last area that needs a quantum leap in both performance and scale, and for this reason, the IP switch-router market recently has seen a great deal of innovation-and service-provider interest. Increases in capacity, improved scalability, and support for better quality of service (QoS) will enable service providers to light up and manage the huge amount of bandwidth they've deployed over the past few years.
IP switch routers or core routers aggregate traffic coming in from the edge at lower-speed interfaces, such as STM-1 or ATM/OC-3, into large trunks running anywhere from 2.5 Gbits/sec to 10 Gbits/sec.
"A lot of people think of the IP switch routers as being in the core, but there is actually a core underneath that," explains Robert Redford, senior director of marketing at Cisco Systems Inc. (San Jose, CA). "There's long-haul transport that actually connects it together. When we talk about an IP core, people think about everything going to Missouri or somewhere in the center of the U.S., when in fact, what happens is an IP backbone is distributed. You'll have a switch router in Chicago, you'll have one in New York, and Denver and Dallas and San Jose."
There used to be a greater distinction between switches and routers. A switch forwarded traffic, in a simplistic way, only at Layer 2. It didn't manipulate the traffic in any way and forwarded it very quickly. Routers, by contrast, were CPU-based and therefore more intelligent devices. They operated deeper inside the packet, at Layer 3.
"If you were a switch, you were fast but dumb, and if you were a router, you were slow but smart. You traded off smarts for performance," explains McLean.
Around the 1995 to 1996 timeframe, the industry started working on integrating intelligence, the routing functionality, into the hardware-beginning with products aimed at the enterprise. The first companies to integrate routing into hardware were the Gigabit Ethernet players-Extreme, Foundry, and Rapid City. Moving the routing into hardware alleviated the dialectic between performance and intelligence, says McLean. It was not until 1999, however, that the industry began to develop hardware-based routing dedicated to the service provider market.
Today's service providers are struggling to get some kind of return on their investment, says McLean. "While 1999-2000 saw sort of breakneck-paced infrastructure build-out, it was all just about pipes, pipes, pipes, pipes, and pipes." Now, they must figure out a way to light and manage that bandwidth-a challenge Redford calls "the services bandwidth barrier."
"Take, for example, a Costco or a Sam's Club, any high-volume warehouse store," explains Redford. "When you go there, they give you the biggest shopping cart you've ever seen. They don't want you to buy a couple of little things. They want you to buy huge amounts of volume."
The same is true, he says, for service providers. "They want you to have a lot of bandwidth, because the more bandwidth you have, the more you can consume," he says. The existing pipes cannot handle the current demand for capacity.
To alleviate this problem, vendors are building larger-capacity systems; the trunks between core routers support more bandwidth now than ever before. In the late 1990s, most pipes were scalable to 622 Mbits/sec. Today, vendors are manufacturing pipes four times that size at 2.5 Gbits/sec, or 16 times that size at 10 Gbits/sec, says Andrew Bronson, product marketing manager for core routers at Lucent Technologies (Murray Hill, NJ).
Scalability is another major trend in the switch-router market. "In core IP routing, 'scalable' means something specific," says McLean. "Can you grow the system in multiple, discrete boxes and have it run like a single router and have it grow in a cost-effective way?"
In a traditional, single-chassis router, there are a certain number of ports, and when you run out of those ports, you have to buy another router, explains McLean. You link those routers together, and then you begin to build a system, but each router is still separate from the others.
"Just now, in the past six months or so, we are starting to see a migration toward distributed systems where, when I buy that next piece of sheet metal and roll it up next to the first one and connect them together, they are actually one router," explains McLean. "I didn't have to waste valuable routed ports to connect them together; I can use ports that draw from switching capacity instead of I/O capacity. These are scalable systems."
There are three main reasons why carriers and service providers are looking at a scalable routing architecture, claims Esmeralda Swartz, director of strategic marketing for Avici Systems (Billerica, MA). First, they are trying to reduce the cost of building and operating their networks, and a scalable router enables them to do that more efficiently than a fixed-configuration router.
"The second thing is that they are trying to figure out how to deliver legacy and new IP services over a converged IP network to increase profitability and squeeze as much margin out of their existing services, as well as introduce services faster," explains Swartz. "This ties into the third reason: 'How do I quickly provision services to beat my competitors to market and start to introduce new revenue streams into my business?'"
The demand for scalability, then, is driven by the need to improve the velocity with which a service provider can provision a service and thereby increase its profits and revenue in a more efficient manner than traditional ATM networks. Today's IP core router products also scale interface sizes beyond the capabilities of ATM networks. The largest ATM interface commercially available is 2.5 Gbits. IP networks have already achieved 10-Gbit interfaces and, in the future, will move into the 40-Gbit, 100-Gbit, and even 160-Gbit arena.
"ATM as a technology is having a hard time trying to scale into those ranges," says Bronson. "So the future is bright for IP networks and IP core routers."
While the future may be bright for IP core routers, McLean contends that it is not necessarily hardware-based improvements that will drive the market. "Quite frankly," she says, "the hard and fast technology from the widget standpoint isn't going to be the gaining factor here. It's going to be technology from the back-end perspective-getting it integrated into their operating models, being able to bill for it, etc. That's going to impact them more than getting the widgets in place."
Another major trend in the core router market is the ability for IP core routers to support QoS, which, until the early 1990s, was only done well by ATM-type products. QoS in an IP network, from the core router standpoint, entails making sure that the routers themselves do not propagate undue delay on the millions of packets that transmit through them.
A higher level of QoS is also a key enabler of service-level agreements, which allow service providers to set up different tiers of services, carry them across the network, and bill for them. The majority of first- and second-generation routers took IP packets and passed them through the core as fast as possible, in a process commonly known as "best effort." There were no mechanisms inside the router to guarantee that a particular user received a certain amount of bandwidth. Previous incarnations were also unable to prioritize packets so that customers who paid more for their services would be guaranteed delivery within a certain time.
"Some of these functions are coming along," says McLean. "With the new crop of routers, you have real sophisticated hardware-based classification, policing, scheduling, stuff like that."
Service providers also want differentiated levels of service says Bronson, "because they are trying to bring new applications online at their end points, like video or voice, that have been traditionally done on TDM or ATM networks. Now they want to do that on top of the IP packet network."
The introduction of new services will require a greater level of reliability than has been traditionally available from the IP packet network, admits Bronson. "The reliability requirements for the core IP switch router are changing in that once we start to support voice on that as well, it's going to take it to a different level."
While IP switch-router vendors are certainly cognizant of the tenuous nature of today's market, they are also confident that they play in a segment that will continue to grow.
Market researcher RHK Inc. (San Francisco) estimates that the worldwide market opportunity for core routers is approximately $4.5 billion this year, $8.6 billion in 2002, and $12.7 billion by 2003.
Redford cites another study that estimates 24 million fiber-mi were deployed in the United States alone in 1999-and deployments have grown substantially since then, he says. "So there's a huge amount of dark fiber out in the network now, and the issue is how do [service providers] light it up? Once they do that, they can get a lot of bandwidth available, and once that bandwidth is available, they can really start to ramp up the services on their networks," he concludes.