RPR: Deadlock Ahead?
Companies made significant progress at the last working group meeting in November, they say. But a battle seems to be heating up as the two main camps -- "Gandalf," led by Cisco Systems Inc. (Nasdaq: CSCO) and its supporters; and "Alladin," including Alcatel SA (NYSE: ALA; Paris: CGEP:PA), Nortel Networks Corp. (NYSE/Toronto: NT), Luminous, Lantern, and other vendors -- haggle over traffic management and a packet format definition.
And it appears the two groups still have a long way to go before January, when they plan to release a single draft proposal (see Cisco Clarifies Ring Strategy ). Indeed, the two camps seem to be on the verge of a deadlock.
In a previous article, Light Reading reported that Cisco had made significant concessions in the final proposal it helped to draft for the technical paper in November (see Cisco in U-Turn on RPR). The company compromised on some key issues, which will ultimately result in it redeveloping its ASICs and software to handle the new standard.
But opponents to the Cisco-backed Gandalf proposal say that the networking giant is dragging its feet on the final two points. Cisco flatly denies these assertions, implying that it is the Alladin side that's unwilling to compromise.
Issue 1: Traffic management
The most hotly contested issue is traffic management, for which each side has a pet solution. The Alladin proposal calls for virtual output queuing, in which packets entering a device are buffered until a message from the output port or queue allows them to enter. This is analogous to cars trying to get onto the highway at rush hour; packets must wait at a “stoplight” until there is enough room on the ring to allow them access.
Alladin supporters say that Gandalf doesn’t provide traffic management. Steve Wood, manager of hardware engineering at Cisco, disagrees. He says Gandalf uses a simple fairness scheme to divide the ring bandwidth among stations. And he claims this is a simpler approach than implementing virtual output queuing, which uses a more complex MAC (media access control) scheme without providing significant performance improvement.
“Gandalf puts the power in the hands of the implementer to do whatever they want to do in terms of traffic management,” says Wood. “Let people make their engineering decisions and see what the market selects."
Some argue that Gandalf’s proposal is technically not the best way to handle congestion on a ring.
“I’ve been studying this for ten years,” says Charles F. Barry, co-founder and CTO of Luminous and co-presenter at the Lightspeed session today. “This is what I did my thesis on for my PhD. I know that technologically this way works better.”
Barry says that mathematical and statistical evidence proves that handling congestion at the input device on the ring is faster and more efficient than waiting for Layer 3 mechanisms to detect congestion and then handle it.
Others in the Alladin camp argue it is not just a technical issue, but an expense issue. Leaving virtual output queuing out of the standard would essentially force vendors using the RPR standard to implement MPLS or some other Layer 3 technology to handle traffic management.
This would be expensive, especially for companies like Alcatel, which are building RPR-enabled Sonet add/drop multiplexers (ADMs) to deliver Ethernet services over rings. It might also hurt DWDM vendors that want to use RPR to deliver Ethernet services over lambdas. These devices do not require routing or MPLS functions, and adding them could be costly.
“If traffic management isn’t included in the standard, it could limit the use of RPR to those wishing to connect devices to routers,” says Frederic Thepot, vice chairman of international affairs for the Resilient Packet Ring Alliance and an independent contractor. “But the standard needs to be as open to DWDM or Sonet ADM vendors as it is for the routing vendors.”
But Cisco disagrees. Wood says that Gandalf fits nicely into ADM or DWDM systems. He also points out that Gandalf opponents also tend to criticize Cisco's use of SRP -- spacial reuse protocol, Cisco's proprietary and controversial implementation of RPR (see Cisco Clarifies Ring Strategy ).
”Opponents of SRP and Gandalf continue to propagate the falsehood that only IP can be transported,” he says. “This is a convenient falsehood with the obvious political benefits.”
Issue 2: Packet format
The working group's other main issue in contention is packet format. Alladin supporters are proposing that the standard use the same packet format as Ethernet with an RPR-specific shim or tag added to it. This is similar to MPLS. Using the Ethernet packet format allows vendors to use cheap, off-the-shelf Ethernet parts to build their systems, says Barry of Luminous. It also means they can use currently available Ethernet testing equipment, he adds.
Conversely, Gandalf’s packet format doesn’t use Ethernet framing. Instead it uses a completely new format, which Barry concedes optimizes IP efficiency. In this format, fewer bytes are used, which means there is less overhead in the network, making the network run more efficiently. But Barry argues that because the format is not based on Ethernet, it cannot take advantage of the inexpensive Ethernet components already available, and it requires new testing gear.
Cisco disagrees entirely with these assertions. Wood says that neither Alladin nor Gandalf can be used with unmodified test equipment. But he has every confidence that the test-set vendors can modify their existing equipment for 802.17, just as they did for SRP when it was originally developed.
As for the ability to use standard Ethernet parts in equipment, Wood concedes that it is handy for first prototypes and easier for startups. But he says it is not the right long-term answer.
“The two leading pre-standard RPR implementations both took the approach of a new frame format that was not compatible with Ethernet,” Wood says. “802.17 is defining a new MAC, and that means that a new MAC silicon will be created. Small parts of an 802.3 MAC will be borrowed for the PHY reconciliation layer, but not the whole design.”
Currently, the RPR work is on schedule, according to Mannix O’Connor, executive secretary for the RPR Alliance and marketing executive at Lantern. The timeline calls for the group to have a single first draft completed by January with the ultimate standard being ratified by the spring of 2003.
But if these issues are not worked out in the January meeting, the standard itself could be in jeopardy.
“These two issues will be the focus of the next meeting,” says Thepot. “We will really be in trouble if we cannot come to some agreement there. The industry may lose faith in RPR altogether, if that happens.”
— Marguerite Reardon, Senior Editor, Light Reading