Special Focus: New ways of routing the Internet

Several router start-ups are proposing different ways to combat the same problem: Internet scalability.

With traffic on the Internet doubling every six months, all of the start-ups agree that a new generation of gigabit and terabit routers is required. Yet they differ on the best architectural approaches to achieve gigabit and terabit rates.

Analysts say the different approaches reflect engineering preference rather than proven advantage over other designs.

"They're coming at it from different angles depending on their backgrounds," says Brendan Hanigan, an analyst at Forrester Research, Inc., in Cambridge, Mass. "It's just another way of cracking the same nut."

Avici Systems, Inc. is pitching a terabit router with a switch fabric that models the growth of the Internet. Avici's Terabit Switch Router (TSR) is designed to scale from 600M bit/sec to multiple terabit/sec to accommodate the geometric growth of the 'Net.

"Scalability is our biggest differentiator," says Avici President Surya Panditi.

True, TSR, like other next-generation routers, will use hardware-based routing, forwarding, multicasting and quality of service to maintain gigabit line rates. But where TSR differs from its brethren is in the line cards.

Each TSR line card sports a so-called Direct Connect Fabric (DCF) that serves as a 70G bit/sec router. Twenty of these line cards will fit into a rack-mountable, dual-shelf TSR chassis with a passive backplane.

With DCF, each line card adds 60G bit/sec of switching capacity to TSR. In a fully loaded dual-shelf, TSR will have up to 1.4 terabit/sec of switching capacity. Internet service providers can scale beyond that by adding more TSR shelves and line cards to their equipment racks.

Another player in the terabit router game is Pluris, Inc. Pluris is proposing massively parallel processing to solve the Internet scalability problem.

The Pluris Massively Parallel Router (MPR) is a collection of single-board computers, or processing nodes, attached via a proprietary data interconnect.

Each MPR processing node has enough memory and processing power to route IP packets at OC-12, or 622M bit/sec speeds, says Vadim Antonov, chief technology officer and vice president of engineering at Pluris.

One or more dedicated processing nodes are used for processing routing protocols. Every processing node has a copy of the forwarding table.

The data interconnect is a patent-pending "self-healing" butterfly switch based on multiple multigigabit serial communication lines. Pluris chose a butterfly switch to achieve linear scalability, Antonov says. "Crossbar switches are great for telephony because you have long-lived connections. If you try to use crossbar switches for packets you soon discover that scaling the crossbar kills you," he says.

Pluris will prototype MPR in February and release it in the second half of this year, Antonov says.

Neo Networks, Inc. is another proponent of massively parallel routing. Neo claims that other architectures that employ fast switching fabrics with modular or separate route processors can become bottlenecks by forcing packets to take multiple trips through the switch fabric. They also can leave networks unprotected, says Mark Cree, Neo's vice president of marketing.

StreamProcessor 2400, which ships in the second quarter of this year, is a 16-slot chassis that sports a 512G bit/sec modified crossbar fabric, seven custom Application Specific Integrated Circuits (ASIC) and more than 1,000 Reduced Instruction Set Computing (RISC) processors. The 16 slots can hold four- and eight-port Gigabit Ethernet modules and a four-port, 2.5G bit/sec OC-48 card.

Multiple chassis can be linked to reach terabit rates, Cree says.

The StreamProcessor views incoming packets as a stream of bits that become an instruction set for the massively parallel architecture. Incoming data is parsed and directed to the RISC processors for application and protocol processing.

While one processor handles application and protocol processing for the data stream, another performs look-up, queuing and prioritization for that stream. This massively parallel architecture yields forwarding performance of more than 400 million frame/sec, Neo claims.

One start-up that does not see the need to move to massively parallel routing is Torrent Networking Technologies Corp. Torrent believes that a shared memory switch fabric, coupled with ASICs and a fast route look-up algorithm, suffices.

Torrent's IP9000 router is composed of three elements: the shared memory switch fabric, embodied in eight- and 16-slot chassis; forwarding engines, which are 10/100 Ethernet, Gigabit Ethernet and ATM modules with route look-up and packet classification ASICs on each port; and a route processor, a stack-on general purpose computer for routing table maintenance and system configuration.

IP9000 is designed to maintain a fast path for all IP traffic by taking the route processor out of the forwarding path. The route processor uses a private channel through the switch fabric to update tables on forwarding engines and only issues updates when changes occur.

Maintaining an entire routing table on each IP9000 port instead of caching known routes helps eliminate the slow path altogether, Torrent claims. A full table search is performed for every packet at wire speed. This avoids the non-deterministic performance of cache-based searches and associated uncertainty in packet latency, according to Torrent.

The search results include the output port, next-hop router address and also any stored service profiles to be applied to the packet. With the information obtained from the routing table search, per-port silicon is able to make all the necessary packet modifications, including queuing and priority information, before presenting the packet to the switch fabric for processing.

This helps the IP9000 scale routing to tens of millions of packet/sec, Torrent claims.

Argon Networks, Inc.'s GigaPacket Node router will have a 150G bit/sec ATM switch at its core, port-level route look-up and support for heterogeneous network and data link protocols.

Nexabit Networks is working on a router with a 2.5T bit/sec fabric and 32 OC-48 interfaces. The router will enter beta trials in June, says Gene Wahlberg, senior vice president of sales and marketing at Nexabit.