Why? First, the folks who are beefing up Ethernet for use in telecom networks are catching up. Although the technology might not be as good as RPR, it may be good enough to win the day because it's likely to be a lot less expensive.
Second, development of the RPR standard has been muddled. To make a long story short, Cisco Systems Inc. (Nasdaq: CSCO) could end up establishing its own spatial reuse protocol (SRP) as the dominant technology for handling IP over fiber rings – whether it’s standardized or not.
Time to back up and look at some basics. The big issue here is the age-old quest for a technology that enables service providers to install a single network to carry all types of traffic.
Right now, most service providers end up having to install multi-layer networks. The first layer after the optics is Sonet or its European equivalent, SDH. Its main job is to provide connections over interlinked networks and automatically reroute traffic around failures (see Sonet (Synchronous Optical NETwork) and SDH (Synchronous Digital Hierarchy)).
The second layer comprises telephone switches for carrying voice or ATM switches for handling data. The ATM layer provides logical connections between different bits of equipment, and controls the performance of those connections (see Asynchronous Transfer Mode (ATM)).
On the data side, there’s a third layer of routers and frame relay switches, which funnel traffic onto ATM connections.
This approach simply can't cut it in metro networks. It costs way too much, makes very inefficient use of bandwidth, and can’t cope with requirements for rapid provisioning of services.
Scores of equipment suppliers are developing metro equipment that aims to address these issues, by replacing the Sonet and ATM layers with something else. In general terms, they’re shaking down into two groups. One, the Metro Ethernet Forum, is promoting Ethernet-on-steroids. The other, the Resilient Packet Ring Alliance, is hoping to come up with a standard for a new layer 2 protocol.
New protocol probably means better protocol, but the Ethernet-on-steroids crew has a big market advantage. According to Nan Chen, president of the Metro Ethernet Forum, it plans to preserve Ethernet’s media access control (MAC) layer in whatever it develops --and that means service providers can charge ahead in deploying plain-vanilla Ethernet switches right away, in the expectation that steroid additions will arrive fairly soon. Preserving the Ethernet MAC layer also means that costs ought to be very low, because the Ethernet components industry is already well established.
The RPR gang also faces another challenge. Each vendor in the group has its own proprietary technologies to promote, and it’s tough to come up with a standard that embraces all of them. To make matters worse, some vendors, notably Cisco, want to optimize RPR for carrying data while others want to make it more of a universal protocol for carrying all types of traffic.
"Our customers buy routers," says Mike Takefman, Cisco’s manager of engineering and chairman of the IEEE RPR Working Group. "They are primarily ISPs. I’m not saying that we don’t care about voice traffic. Cisco has been a strong supporter of voice over IP, but most of our customers handle data traffic."
"RPR should be more than just hooking a bunch of routers together," says Nader Vijeh, senior vice president of research and development and CTO of Lantern Communications Inc., a member of a group of vendors proposing a more voice-friendly alternative to SRP for the IEEE 802.17 standard. Others in the group include Luminous Networks Inc., Dynarc, Nortel Networks Corp. (NYSE/Toronto: NT) and Alcatel SA (NYSE: ALA; Paris: CGEP:PA).
Superficially, that’s what is behind the current division in the RPR standards effort, although there’s more to it than that. Cisco has managed to put competitors’ noses out of joint by trying to railroad the whole standardization process within the Institute of Electrical and Electronics Engineers (IEEE).
On one front, Cisco has adopted the same policy that it uses in the Internet Engineering Task Force (IETF) - namely, it's taken its existing technology and argued that SRP should form the basis of the standard because it is already widely deployed (by Cisco) and thus proven in practice. That has irked rivals who don't want to hand the RPR market to Cisco on a plate.
On another front, Cisco has been propagating SRP support in the marketplace, notably by giving away code to semiconductor companies to encourage them to make SRP chip sets. Mindspeed Technologies and Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) have already jumped on the bandwagon. Cisco has acquired a startup semiconductor company called AuroraNetics that is also working on pre-RPR chipsets.
The idea is that the availability of SRP chips will encourage vendors to develop SRP equipment. Riverstone Networks Inc. (Nasdaq: RSTN) and Spirent Communications have already announced products supporting Cisco’s technology.
All of this means that Cisco doesn’t have a lot to lose if the IEEE 802.17 Working Group gets locked into an argument that delays everything. Support for SRP in the marketplace is growing, strengthening Cisco’s hand in the meantime.
While the RPR crew squabble and Cisco pushes ahead with SRP, the Ethernet-on-steroids gang is quickly catching up. The Metro Ethernet Forum already has more members than the RPR Alliance and expects to have draft specs for its telecom-optimized Ethernet by the end of the year. The RPR working group isn't even supposed to have a draft proposal completed until the January meeting. Members will then be able to review and make changes to the draft until it is finally voted on next summer.
One of the key issues being tackled by the Ethernet-on-steroids gang is the time it takes to restore service in the event of a line or equipment failure. In a traditional Ethernet network, the packets must use a spanning tree algorithm to find an alternate path. This can take 5 to 10 seconds – several orders of magnitude slower than the 50 millisecond restoration times supported by Sonet.
Bob Klessig, vice president of Telseon Inc., an Ethernet service provider and co-chairman of the technical committee for the Metro Ethernet Forum, says that faster spanning tree algorithms have been developed to get restoration times down to 1 second.
"The 5 to 10 second fail-over will be transparent to most data applications," he says. "For voice applications there has been a lot of work to get the restoration times down much further. Even with 1 second restoration, it would only sound like a click in your telephone call."
In Nan Chen’s terms, this amounts to "the cheapest, good enough" technology for metro networks. He points out that Ethernet beat out other LAN technologies for exactly the same reason. It wasn't better than token ring, FDDI and ATM-to-the-desktop, it was simply cheaper and good enough.
In essence, the RPR crew is trying to leverage its way into a battle that has already begun between Ethernet and Sonet, both of which are being improved to address metro requirements.
"It’s hard to say if there will be a need for RPR," says Geoff Bennett, vice president of technology advocacy for Marconi PLC (Nasdaq/London: MONI). "By the time there is a standard, Cisco will have flooded the market with SRP and the standard may not be interoperable with Cisco’s installed base. New developments in Metro Ethernet and dynamic Sonet will also be out by then."
- Marguerite Reardon, Senior Editor, and Peter Heywood, Founding Editor, Light Reading