The IEEE standard for Resilient Packet Ring may be closer to reality after the submission of a new proposal that evolved from compromises made by Cisco Systems Inc.- and Nortel Networks Corp.-led factions.
Dubbed Darwin, the proposal was to be voted on by an IEEE 802.17 working group Jan. 25 (after press time). If approved, the proposal will have to work its way through an IEEE standards process that likely won’t produce a ratified standard for about another year, although observers say it should be safe for carriers to start installing prestandard equipment and upgrade the software or hardware later.
RPR has gained attention among carriers and equipment suppliers looking for a better way to offer high-speed Ethernet-based services across metropolitan-area networks. RPR, which will be supported in switches and other gear, is intended to optimize metropolitan ring topologies for packet transport with resiliency matching or exceeding SONET’s.
The new Darwin proposal incorporates the best of Cisco’s and Nortel’s plans with the intent of reaching a broader consensus, according to John Hawkins, a Nortel marketing manager. While the original proposals haven’t been yanked off the table, it’s clear from discussions with Cisco and Nortel officials that Darwin has supplanted their respective Gandalf and Alladin pitches. Slews of other companies, including Appian Communications Inc., Intel Corp. and Riverstone Networks Inc., have also thrown their support behind Darwin.
The changes the group made in developing Darwin focus on two areas: bandwidth management and failure protection.
“Within this algorithm, there are ways to adjust network performance depending on the amount of congestion in the network at any given time,” Hawkins says. “This allows network operators to be more proactive about network management.”
According to Hawkins, the Gandalf proposal uses a management algorithm that only adjusts network performance when congestion starts. The Alladin proposal uses what Hawkins calls an “avoidance method,” which is a more passive means of network monitoring.
Raj Sharma, director of technology strategy for optical network equipment maker Luminous Networks Inc., says Darwin’s bandwidth management mechanisms let carriers offer and enforce different qualities of service.
The Darwin proposal incorporates two separate means for protection against network failure called wrapping and steering.
The wrapping method, which is required under the proposal, occurs when two links on a ring become aware that a neighbouring link is failing and then “wrap” the traffic around the failed link unbeknownst to the other links on the ring.
Steering, which is optional, happens when all nodes on a ring detect failure and involves injecting packets into the ring that are then routed around the failure.
Jeff Baher, a Cisco senior marketing manager, says that both methods of protection can restore the network in 50 msec, like the current, industry-accepted SONET recovery times.
“There is just a lot of religion and philosophy around both methods, and Darwin allows the vendor to select which type of box he wants to have in his network,” he says.
Sharma says the two protection methods also allow for backward compliance and simple implementation.
Work on Darwin began in July when several working group members saw the need to merge two different worlds of thought.
“The first step was to get the two gorillas to shake hands,” Sharma says. “We needed to work this out and come together. There is a real business issue here, and Darwin has taken the vision of RPR and reaffirmed it.”
Gandalf is based on a controversial Cisco-based technology called Spatial Reuse Protocol (SRP). In past meetings, backers of Nortel’s Alladin proposal argued that Gandalf didn’t provide any traffic management capabilities, but Cisco argued the technology uses a fairness scheme to divide the ring’s bandwidth among nodes. Alladin supporters also argued that Cisco was dragging its feet to make its technology the de facto standard.
Cisco responded that Alladin supporters were unwilling to compromise.
