ACCESS TECHNOLOGY RPR—Building A Better ? Robert Bellman

Will metro carriers go for services.” If I were a member of the 802.17 work- resilient packet ring ing group, the breadth of that statement would worry me. The working group needs to keep its technologies? Will vendors eye on the ball—efficient, robust metro-area data be able to agree on services—or 802.17 is destined for a one-way trip standards? to the protocol museum. What Goes Around May Not Come Around etwork service providers need a new kind Service providers like their existing SONET rings of optical ring technology to deliver high- —and for good reasons. Besides SONET’s 50- speed services—at least millisecond restoral, the rings reach more cus- N that’s what some vendors think. Compa- tomers with less fiber than other topologies and nies like Appian Communications, Cisco, Dynarc require fewer switch ports at busy hub sites. But and Lantern Communications are developing SONET, which was designed for voice circuits, resilient packet ring (RPR) technologies, which wastes bandwidth when carrying packet traffic, replace Ethernet’s media access control (MAC) and the spanning tree algorithm that guides most layer with a new ring-based MAC. Atrica and a Ethernet switches purposely breaks rings in order few others are addressing the challenge with ring- to prevent bridging loops. based solutions that preserve the Ethernet MAC. Packet over SONET (POS) offers a partial All these approaches promise service providers solution, but only for point-to-point links. A better the best of both worlds: Ethernet’s low cost and solution, according to the RPR proponents, would packet data efficiency, plus SONET’s ring struc- be a new packet MAC that uses rings efficiently ture, reliability and rapid restoral. but also exhibits the resilience and QOS of Sensing a groundswell, the IEEE formed the SONET. 802.17 working group late last year to standardize The 802.17 working group plans to have a first RPR. A marketing consortium, the RPR Alliance, draft standard in January 2002, with final techni- was formed in January. The IETF also has a work- cal changes by September 2002 and official stan- ing group called IP over RPR (IPoRPR). dardization by March 2003. A healthy mix of ser- But 802.17 has a sweeping agenda that could vice providers and equipment manufacturers slow progress toward its most basic goals. A attends the working group meetings. As usual, recent RPR Alliance press release declares, “RPR each equipment manufacturer is lobbying for a will support carrier-class, service-level-agree- standard that fits its technical vision (and mini- ment-based metro Ethernet, IP, and legacy TDM mizes its re-engineering). The working group has

FIGURE 1 Destination stripping

Bob Bellman is president of Brook Trail Research (www.brooktrail.com), a consultancy specializing in network technology assessment 1a Source stripping 1b Destination stripping and marketing.

40 BUSINESS COMMUNICATIONS REVIEW / SEPT 2001 Use BCR’s Acronym Directory at www.bcr.com/bcrmag used these inputs to build consensus for high-level Support for multiple service types: RPR will objectives before diving into the technical details. adapt to future requirements. These objectives, as reported by the RPR Vendor interoperability: RPR equipment Alliance, comprise a rough outline of the architec- from different vendors will interoperate on the Some metro-area ture, commitments to motherhood and apple pie, same ring. and some noticeable opportunities to stray from In other words, RPR’s primary mission is to Ethernet vendors the basic mission: make optical rings more efficient for packet traf- are opposed to Dual counter-rotating ring topology: In dual- fic, but the standardization effort has attracted ring topologies, SONET uses only one ring to some developers who want to see it do more, the new RPR MAC carry live traffic; the other is reserved as a backup. especially in the realm of QOS and traffic control. To increase fiber utilization, RPR will send traffic Others ask why Layer 3 mechanisms (such as over both rings (in opposite directions) during DiffServ and MPLS) can’t be used to provide normal operation. these functions. A fully distributed access method without a master node: An RPR ring will continue operat- If I Had a Hammer ing despite the loss of any node. Another opportunity to go astray is represented in Protection switching in less than 50 millisec- the 802.17 working group’s mission statement: onds: In the event of a fiber break or node failure, “The IEEE 802.17 Resilient Packet Ring Working RPR will restore service at least as fast as Group will define a Resilient Packet Ring Access SONET. This will likely entail linking the two Protocol for use in Local, Metropolitan and Wide rings into one, so some traffic may get bumped. Area Networks for transfer of data packets at rates Destination stripping of unicast traffic: In scalable to many gigabits per second.” some older packet ring architectures, the source The last protocol that tried to cover “Local, node removes unicast packets after they come all Metropolitan and Wide Area Networks” was the way around (Figure 1a). With RPR, destina- ATM. Untold millions of dollars were wasted try- tion nodes remove their unicast packets, freeing ing to make ATM, which was initially conceived downstream bandwidth for reuse by other flows as a wide-area technology, suitable for local-area (Figure 1b). Together with packet multiplexing networks. and counter-rotating rings, destination stripping Is 802.17 making a similar mistake? Bob Love, more than doubles RPR’s total throughput com- Chair of the RPR Alliance and Vice-Chair of the pared to SONET. 802.17 working group, thinks not. “If you have a Support for multicast traffic: Multicast pack- hammer, the world looks like a nail,” said Love, ets travel once around the ring to reach every “but it’s clear that MANs are the sweet spot.” For node. In contrast, mesh networks must replicate the sake of RPR’s original mission, let’s hope the multicast packets in order to reach all destinations. rest of the working group agrees. Support for up to 10 Gbps: RPR will be fast More substantive concerns center around the enough to carry Gigabit and 10-, protocol “sandwich” RPR will create. Users will but will also support lower data rates. see an Ethernet service interface that plugs direct- Support for both SONET/SDH PHY (physi- ly into their routers and switches. Underneath that cal layer) and the 1- and 10-Gbps Ethernet interface, the RPR MAC will move packets PHYs: Support for existing PHYs will let RPR around the ring, while underneath the MAC, an products use widely available components. Ethernet PHY formats the bits onto the fiber. PHY and payload agnostic: To be truly uni- Some metro-area Ethernet developers are versal, the RPR MAC will be independent of the opposed to the new MAC altogether, including PHY layer and will not interfere with customer David Yates, Atrica’s marketing VP. “RPR is as payloads. similar to Ethernet as is to Ethernet; Plug-and-play support: New nodes may join that is, they’re not similar at all,” he said. “Ether- the ring without manual configuration. net is standard, understood and based on cheap Managed objects: By defining managed components that scale fast. RPR will need differ- objects, the RPR standard will facilitate OSS inte- ent components, different management, and so gration. on.” He asserted that RPR-based Ethernet will Support for services that require bounded cost more than ordinary Ethernet. delay and jitter and guaranteed bandwidth: Atrica takes a different approach with its fami- RPR will deliver TDM-like QOS. Welcome to the ly of Optical Ethernet Switches. It uses a 10-Gbps top of the slippery slope. If RPR strays too far Ethernet MAC over WDM wavelengths and relies from Ethernet in attempting to support TDM, it on existing standards to make up for Ethernet’s won’t be able to use commodity chips, and it may shortcomings. For example, DiffServ and 802.1p pose more direct competition to SONET than let switches manage traffic priorities, while Multi- some suppliers would like. protocol Label Switching (MPLS) provides fast Dynamic weighted bandwidth distribution: recovery from outages. RPR will allocate bandwidth to competing flows Meanwhile, the folks at Appian aren’t entirely on demand. opposed to an RPR MAC, but they also are con-

BUSINESS COMMUNICATIONS REVIEW / SEPT 2001 41 cerned about its complexity. Anand Parikh, Appi- and those already on the ring (Figure 2). Should an’s vice president of product marketing and busi- transit packets take precedence over entering ness development, agrees that matters like fairness packets, or should all packets compete equally for Cisco’s Spatial and QOS should be handled by existing standards bandwidth at every hop? So far, the 802.17 work- outside of RPR. “Keep the architecture simple,” ing group has not decided. Reuse Protocol he said. “Otherwise, nothing will get done. RPR With Cisco’s SRP, transit packets take prece- meets many RPR should use existing standards where possible. dence over entering packets, so there’s no packet There’s no need to reinvent the wheel.” loss on the ring itself. “This is different from how objectives Appian is also concerned that a complex MAC Ethernet switches work,” Baher pointed out. Ordi- layer will pollute the PHY layer and make low- nary Ethernet switches, which lack a cut-through cost Ethernet framers unusable. If RPR can’t path for transit traffic, exhibit varying packet loss deliver Ethernet services at Ethernet prices (or throughout the network as traffic congests at each better), who will pay for the equipment? node. If the 802.17 working group decides against transit cut-through, then RPR will lose a potential Yield To Oncoming Traffic? advantage over ordinary Ethernet switching. By contrast, Cisco doesn’t share Appian’s or Atri- Transit traffic should take priority, agrees Bob ca’s concerns. In fact, it already has a new MAC Schiff, senior director of strategic marketing at layer, the Spatial Reuse Protocol (SRP), that Lantern Communications, and that should be part meets many of the RPR objectives. SRP is used in of the standard. “You must protect the traffic that’s Cisco’s Dynamic Packet Transport (DPT) product already in the ring,” he said. “It’s like a traffic line, which has been shipping for more than two rotary. Once you’re in the rotary, you’re okay. The years, largely to service operators. only challenge should be insertion delay.” (I guess Cisco submitted SRP to the IETF in 1999 as infor- Schiff hasn’t driven in Boston lately. We acceler- mational RFC 2892, and participates in both the ate into rotaries.) Lantern’s Metro Packet Switch IEEE working group and the RPR Alliance. products, which employ a cut-through scheme, are Cisco’s first DPT products were router inter- scheduled for field trials this fall. faces, and the initial applications were for ISP intra-POP rings. “We got a head start,” said Jeff SONET Or Simple? Baher, a senior marketing manager at Cisco, Another controversial topic is the relationship “before metro Ethernet services were such a big between RPR and TDM. At stake are not only the deal.” Compared to an ATM or POS mesh, DPT definition and scope of RPR, but also the migra- gave service providers a way to keep OC-3 port tion path from SONET. “There are two camps,” counts low, and to minimize the number of IP sub- said Cisco’s Baher. “One sees RPR as a SONET nets within a site. From there, DPT spread to inter- ADM replacement. The other sees RPR as simply POP connections and use by cable modem opera- a high-performance packet MAC.” tors to connect head-ends with data centers. Now The rationale of the SONET replacement camp Cisco boasts more than 12,000 installed DPT echoes every convergence vendor’s slideware: ports at more than 160 customers, and Baher Voice is a big revenue producer, so RPR has to expects DPT “to migrate to our high-end Ethernet handle voice efficiently. In contrast, the keep-it- switches, especially our IP-oriented ones.” simple camp argues that TDM support will be When planning that migration, however, Baher RPR’s undoing. is concerned about where 802.17 compliance “There’s no need to build the MAC of all might lead. In particular, Baher questions the rela- MACs,” insisted Baher, “no need to pull out all tionship between packets entering the RPR ring the SONET gear. RPR is okay for IP phones, but not for TDM per se. If RPR tries to do everything, FIGURE 2 Competition For Ring Access it won’t do enough TDM for TDM folks, and it won’t do enough packet for packet folks. Don’t kill RPR by overloading it.” RPR Appian’s marketing VP, Karen Barton, took Node the simplicity argument one step further. “It’s about today’s reality,” she said. “RPR needs to co- exist with traditional SONET mechanisms. When TDM is fully packetized, the need for SONET will go away. But how many big service providers have accepted packets for carrying voice or pri- Entering vate line?” Packets Eastbound Fiber Appian, therefore, offers a hybrid approach. Its Optical Service Activation Platforms (OSAPs) Transit dedicate some channels on a SONET ring to pack- Packets et traffic and leave others for ordinary TDM. This Westbound Fiber allows OSAPs to share fiber rings with conven- Ring Interface tional SONET ADMs.

42 BUSINESS COMMUNICATIONS REVIEW / SEPT 2001 FIGURE 3 Atrica Optical Ethernet Switches Ethernet

Ethernet SONET SONET TDM ADM

SONET TDM SONET ADM

SONET SONET TDM ADM

Ethernet

Dynarc’s implementation is more a blend than The Ghost Of Protocols Past a hybrid. Dynarc is headquartered in Sweden, The diversity of Ethernet/TDM co-existence tech- where “networks are a couple of years ahead of niques should alert the 802.17 working group to the U.S.,” according to Fredrik Hanell, the com- proceed with caution. Equipment vendors are like- pany’s vice president of marketing. Dynarc is cur- ly to ignore any standard that requires massive re- rently supporting a project in Stockholm that is engineering to support TDM, especially since car- bringing 100-Mbps Ethernet-based network riers are mostly happy to run their voice circuits access to 20,000 apartments. over SONET. Dynarc’s Channelized Reserved Services Likewise, if technical complexity makes RPR (CRS) architecture integrates IP with Dynamic too expensive, some other flavor of optical Ether- synchronous Transfer Mode (DTM), an ETSI net will capture the market. In short, if the work- standard for on-demand circuit switching. CRS ing group tries to be all things to all carriers, stan- routers carry everything in IP packets and group dards-based RPR products will never make it past the packets into channels. “A channel is a logical the interoperability demos. resource defined by anything in an IP packet,” Networking history already has at least three explained Hanell, “source address, destination defunct packet ring standards—FDDI, DQDB and address, QOS bits, whatever.” token ring. And who even remembers failed con- Channels, in turn, are mapped into TDM-like vergence technologies like FDDI II and Iso-Ether- timeslots. Channel capacity can be fixed, perhaps net? To avoid membership in this unfortunate to tunnel a T1 through the network, or flexible to club, the 802.17 working group should stick to its accommodate less sensitive flows. One DTM fea- knitting—efficient, robust, metro-area data trans- ture that Hanell would like to see in RPR is the port—and leave grander visions for another day ability to expand ring capacity by turning on WDM wavelengths. Yet another variation on packet/circuit coexis- Companies Mentioned In This Article tence is offered by Atrica’s family of Optical Eth- Appian Communications ernet Switches. As mentioned above, Atrica (www.appiancom.com) rejects the RPR MAC in favor of a 10-Gbps Eth- Atrica (www.atrica.com) ernet MAC over WDM wavelengths. Moreover, Atrica’s Optical Ethernet Switches carry low- Cisco (www.cisco.com) speed TDM traffic in Ethernet frames. At speeds Dynarc (www.dynarc.com) from T1 (1.5 Mbps) to OC-12 (622 Mbps), Atrica IEEE 802.17 Working Group uses Ethernet circuit emulation. At those speeds, (www.ieee802.org/17/) serialization delay is not a problem, according to Lantern Communications David Yates. At OC-48 and up, however, SONET (www.lanterncom.com) flows get their own wavelengths alongside the RPR Alliance (www.rpralliance.com) Ethernet lambdas (Figure 3).

BUSINESS COMMUNICATIONS REVIEW / SEPT 2001 43