New users and hungry applications are weighing heavily on the network. By some accounts, bandwidth demand is multiplying by three to four times per month. But new routers holding up network backbones are bulking up to carry the load.
Vendors are fighting over who has the biggest, baddest router of them all. Application specific integrated circuits, or ASICs, which put basic processing power in the hardware as opposed to the software, are enabling all these routers to pump packets through the network at heart-stopping rates. Beyond the bulk, core router vendors are focusing on network management, reliability, security, and real-time communications and billing as they tip the scales.
Companies like Metromedia Fiber Network Inc. (www.mmfn.com) and its data center/Internet connectivity subsidiary AboveNet Communications are installing core routers such as Juniper Networks Inc.'s (www.juniper.net) M160 to pump up backbone capacity. MFN is installing 170 of the M160s in its Internet backbone as part of a global OC-192c (10gbps) upgrade. In addition to increased bandwidth capacity and processing power, the M160 routers bring new benefits in network security and network traffic shaping. The routers' Internet II processor ASIC filters 10-gigabit links at wire speed, enabling increased security measures against various forms of denial of service attacks. "Our purchase of the M160 routers represents a valuable addition to our suite of defenses against attacks on our network," says Nick Tanzi, president and COO at MFN.
By 2004, studies say some PoPs will be handling 10 petabits (Pb) of traffic, according to Claudio Mazzuca, a product manager at Hyperchip Inc. (www.hyperchip.com), claims to have "the most scalable router on the horizon." About 20 percent of that traffic will be packet switched while the other 80 percent will be optically switched, he says. Twenty percent means 2 Pb in a given switching office, and at 2 Pb a carrier would need 50,000 OC-768 ports, or 40 gigabits, he continues.
"At 50,000 ports, Hyperchip lets you reach to that level," says Mazzuca. "That's with one routing table, and configuration management and network management on one system. With other systems you have to add routers and interconnect with SONET interfaces--they won't be able to scale in a nonblocking fashion. That's a cluster of routers, and the problem is you use lots of interface ports."
Carriers want to terminate as many interfaces on a single box as possible so they only have one box to purchase and one box to configure, says Ram Krishnan, director of internetworking for the IP core routing group at Lucent Technologies Inc. (www.lucent.com).
Get It, You'll Need It
"There is a real need for high capacity," Krishnan says. "If you don't need it today, you'll need it about a year from now and want to be able to use existing equipment" rather than run into problems with obsolescence.
Lucent this summer announced the NX32000 Multi-Terabit Switch Router for secondary metropolitan markets, which offers 3.2 terabits of total capacity. It had previously come out with a larger router, the NX64000, with twice that capacity. According to Lucent, the NX32000 offers up to 100 times faster delay performance compared to other IP multiterabit router products, enabling real-time service support over a native IP backbone and the lowest power consumption per OC-48c port, offering three times better power efficiency over other terabit router platforms. The product supports interfaces from 45mbps (DS-3) to 10gbps (OC-192), directly connects with optical networking systems at OC-192 or 10gbps rates, and integrates DWDM optical components directly into the product. And the NX32000 provides multiservice capability supporting frame relay, IP and MPLS in a single box from a single line card and offers line rate QoS support for real-time applications.
And, oh, by the way, Krishnan says these routers are the fastest in the industry.
"We have the first release [of] 16 OC-192s; no other platform can support that in one chassis," he says, explaining it can support 6.4 terabits of capacity in one switch fabric. Cisco Systems Inc. (www.cisco.com) has its 12012, which can support 11 OC-48s. And Cisco has talked about an even larger router called the 12016, but has not yet made that generally available, Krishnan told xchange during an interview in late July.
Outside the Box
According to Vijay Parikh, vice president and general manager of the IP PoP systems business unit at Cisco, "It's about networks, not about boxes," explaining that network management is a key part of its core routing strategy.
At SUPERCOMM in June Cisco introduced an element manager, which manages its GSR12000 high-end router family. The manager includes a GUI for easy network element provisioning and to help upgrade line cards. A software module called the Cisco Provisioning Center helps customers provision services on the routers such as VPNs. Another software product, the Cisco Assurance Center, detects faults with particular customer's services, a type of service, link or whatever. "Like if your VPN service has a problem, you can see what customers are affected," says Parikh.
"So our whole management story is a fairly comprehensive one," he adds.
David Skirmont, a systems architect at core router vendor Pluris Inc. (www.pluris.com), says his company, Cisco and Avici Systems Inc. (www.avici.com) all have new products that enable a carrier to connect multiple chassis and manage them as one router.
Pluris by the end of the year with begin betas for its Teraplex 20, which offers 16 line/control card slots, 140 gigabits of line capacity, 14 OC-192s per chassis and scales up to 128 chassis.
"We have a fiber optic interconnect between the chassis that gives you a lower weight, and the ability to put chassis up to 100 meters away [since] in a CO environment you don't necessarily have contiguous space."
Pete Chadwick, vice president and product management at Avici Systems also chants the scalability, reliability mantra in describing his company's router strategy. Avici's core router has redundancy built into the switching fabric, so cards can be upgraded--and recognized by the network--without having to reboot the router. That router can support up to 40 OC-48s in a 7-foot chassis and can expand across multiple chassis for 560 OC-48s, he says.
Split Decisions
As mentioned earlier, ASICs allow these routers to scale to these new heights. But ASICs, known for their brawn, may also get some brains going forward, says Arun Jain, director of marketing for IP infras-tructure at Nortel Networks Corp. (www.nortelnetworks.com), which resells Juniper's routers.
"In the old routers, every decision was made in software. They split that up so forwarding happens in ASICs and the basic routing--finding where to go--is in the software. Therefore the routers become faster," says Jain. "That's already taken place."
The next trend is IP billing, according to Jain. But if carriers are going to do IP billing, they need ASICs that can count packets. Counting packets can enable a carrier to generate reports so it can assure users it's meeting its SLAs. Next-generation ASICs, called packet processors, will do that, he says.
