Cisco Systems Inc. and Nortel Networks devised incompatible resilient packet ring (RPR) schemes more than a year ago and have shipped equipment into live networks. Meanwhile, startups such as Lantern Communications Inc. and Luminous Networks Inc. have proposals of their own on the table. The 802.17 group faces the challenge of boiling down all the ideas into a single, 12-part RPR proposal scheduled to go to ballot in the spring.

Among the hurdles already cleared was some participants' mistrust of Cisco. Multiple sources said that Cisco made an early play to push its Dynamic Packet Transport (DPT) as the model for RPR, but that the routing giant had to back down against tough opposition. Cisco officials told EE Times they're willing to work by consensus and aren't trying to force DPT on the committee.

"Our goal was to develop a solution to a problem and then open that up to a standards body," said Rob Redford, vice president of marketing for Cisco's public-carrier IP group.

Meanwhile, skepticism continues to brew outside the RPR camp, as some carriers and equipment providers question whether RPR will be able to achieve its goals of low cost and Sonet-like restoration. Lucent Technologies Inc. and others have begun proposing alternatives, although whether they can gain ground against the RPR effort is yet to be seen.

Battle of titans

Resilient packet rings are a way of sending traffic on physical rings — the topology most commonly used in metropolitan areas — while preserving the reliability of Sonet but discarding its inefficiencies. Where Sonet keeps a backup ring free in case of catastrophe, RPR would use the backup ring in addition to the primary ring, theoretically doubling the available bandwidth. And where Sonet forces all traffic to traverse the entire ring, terminating at the source node, RPR would let packets travel only the arc between the source and destination nodes.

The standard boils down to a new type of media-access controller (MAC) that decides if traffic should be accepted into the node or forwarded farther along the ring. Ideally, this MAC would be able to handle both Sonet and Ethernet, accommodating the needs of old and new carriers, but would come with the low-cost benefits of Ethernet.

Two key candidates for the standard are Cisco's Dynamic Packet Transport (DPT) and Nortel's Optera Packet Edge, once called Inter-WAN Packet Transport.

Cisco developed DPT to make the Internet Protocol (IP) run more efficiently on metro rings. IP routing is necessary, because RPR will let carriers offer profitable services, Cisco's Redford said. Nortel, on the other hand, considers its model to be better geared for edge networks that don't use routers. The goal of Optera Packet Edge was to create Layer 2 switching for metro rings, providing an inexpensive alternative to Sonet transport.

Cisco claims Nortel's emphasis on transport won't make carriers any money, a sore spot as profits dry up in the wake of the telecom collapse. Nortel points out that carriers don't necessarily want the complexities of routers. "Carriers tell us they don't have router expertise in their edge networks," said Gary Southwell, director of technical marketing for Nortel's Optical Ethernet Division. Nortel crafted Optera Packet Edge to be implemented by technicians accustomed to Layer 2 switches.

On Cisco's side, Redford said the point isn't routing for routing's sake, but rather to give service providers the chance to do more than transport bits. IP-based services such as virtual private networks (VPNs) are a potential moneymaker. And VPNs need IP, he said, because it allows packets to be differentiated so that customers can be billed accordingly.

Nortel's scheme, he said, relies on a point-to-point arrangement that doesn't scale as well as IP does. "If we could build the Internet with point-to-point connections, we would have done it already," Redford said. "It's the same old issue: Layer 2 networks aren't scalable."

30 percent load common

But Southwell responded that routers, aside from being more expensive than Layer 2 switches, don't use all their available brainpower; a CPU load of 30 percent is common if latency-free routing is to be maintained, he said. "So it's not just about bandwidth, it's about how fast packets traverse the box."

The protocols even differ in how they use the MAC. Nortel's method allows the MAC to view a fraction of each packet header, while Cisco's stores the entire packet header in memory before deciding whether to forward it on, Southwell said.

Many of the 78 companies involved in the 802.17 meetings see room for improvement in both companies' approaches. Raj Sharma, director of product-line management for Luminous, agreed that Cisco's approach adds the undesirable expense of routing to RPR. But he pointed out that Nortel's Optera Packet Edge retains the idea of Sonet's time-division multiplexing (TDM), inserting packets into available time slots around the ring. Many companies want RPR to avoid TDM because of its inefficiency — it requires that every node keep track of what time slots are available on the network.

Startup fever

Some startups, such as Luminous, are shipping their own versions of RPR already, and many startups involved in the 802.17 work have added proposals to the specification. For example, Luminous has suggested that nodes be in contact with one another to communicate line breaks and to avoid congestion. Nortel and Cisco, by contrast, allow the nodes to discover this on their own as traffic retreats from a line break or backs up due to congestion, arguing that the extra time doesn't affect the overall flow of the network.

Another area that could be ripe for debate at the working group meeting next week is the issue of fairness during network congestion. "Fairness" is something of a misnomer for a feature that makes it possible to give higher-paying customers their guaranteed bandwidth during difficult times; vendors jokingly call it an "unfairness" protocol. Companies including Lantern have proposed varied algorithms and approaches to apply to the fairness issue.

Yet another question lies in the classes of service — tiers representing the importance of traffic streams — that RPR would offer. Cisco's DPT handles only two classes; other proposals would boost the number to eight.

So far, the standards process has remained on schedule. "We've got very good traction," Sharma said, noting that 70 to 80 percent of the draft proposal is completed.

Sources say that the participants are getting along, too — a big step considering the early friction reported with Cisco. According to Nortel's Southwell, Cisco's push to model RPR after DPT prompted a group of startups including Lantern and Luminous to ally against what they considered a takeover attempt, and to enlist Nortel's help.

"Suddenly there was a consortium behind our approach, because the small guys were afraid Cisco would get its way," Southwell said.

In a way, Cisco deserves credit for starting 802.17. The company presented DPT to an IEEE planning session in 1999. When the speaker asked if any vendors were doing something similar, "about 14 or 15 different companies raised their hands," Sharma said. After that, preparatory work on 802.17 began.

Cisco executives denied an interest in taking over the proceedings and said it's natural for each company to suggest its own technology for standardization. "If anybody has gone to those working-group meetings, they know you can't shove something through," Redford said.

It didn't help that Cisco announced the acquisition of AuroraNetics Inc. (San Jose) in July, during the last set of RPR meetings. The chip startup was developing technology to enhance Cisco's Spatial Reuse Protocol, a critical piece of DPT. This touched off fears that Cisco was trying to acquire the necessary RPR pieces in hopes of dominating the standards negotiations.

Cisco denies any land grab with AuroraNetics. "We are committed to the standards process. We are committed to licensing the AuroraNetics technology," Redford said.

Sharma of Luminous, meanwhile, said that Cisco has turned down the heat. "They've admitted RPR has to have more oomph in the market" than DPT alone can provide, he said.

Other options

Though RPR has a high profile, not all carriers and equipment vendors are convinced it is the best bet for metro rings. Many compare it unfavorably to token ring, an older protocol in which nodes on a ring had to take possession of a "token" to get permission to transmit.

Equipment vendor Atrica Inc. (Santa Clara, Calif.) has been particularly vocal about finding alternatives. And in July, Lucent tipped an Ethernet-RPR hybrid that would not require a specialty MAC, and chip startup Luxpath Networks (Dublin, Ohio) presented an alternative way to provide Ethernet restoration on rings.

Among the concerns is that RPR won't come cheap. The standard piggybacks off the Ethernet frame format, but the special MAC hardware required won't gain from Ethernet's cost reductions, said Jay Gill, director of IP service products at Telseon, the Englewood, Colo.-based metro network provider.

"It's likely to require new chip hardware to implement, and that means you won't have the full price advantage you get with Ethernet," Gill said.

Telseon, for its part, sidestepped RPR in favor of the Rapid Spanning Tree Protocol (RSTP), which was closer to being standardized at the time Telseon was designing its network. "We put our eggs in that RSTP basket knowing that the standard was becoming solid," Gill said. "Two years later, RSTP is a done deal. By contrast, RPR has come along as a standards-body effort much later."