RPR Soups Up Sonet and Ethernet

RPR (Resilient Packet Ring) can help transform that Studebaker into a Ferrari by accentuating Sonet's and Ethernet's positives and attenuating their negatives. Although RPR was designed for carrier, metropolitan and campus networks, it can be implemented anywhere resiliency and efficient bandwidth usage are needed. RPR is not meant to replace Sonet but to ride on it and add value to it. Sonet will remain the king for long-haul networks.

And though carriers will be the first use RPR, primarily because of the cost savings, vendors are already designing RPR-based products for the enterprise. For example, AT&T has announced its Managed OptEring Service, which runs over an RPR network.

Looks Can Be Deceiving

RPR looks like a Layer 1 technology, but it's a Layer 2 implementation designed to run on Layer 1 standards such as Sonet/SDH. Although RPR can operate over Gigabit or 10 Gigabit Ethernet, Corrigent Systems and other vendors say they prefer using Sonet as Layer 1.

The reason for developing RPR as a Layer 2 protocol was simple: cost. Because RPR runs on top of existing technologies, carriers, service providers and enterprise customers don't need to scrap expensive network infrastructures--they can just add RPR equipment to the mix. And RPR traffic can travel on Sonet/SDH networks using Sonet VTs, so RPR can be added to networks without affecting current Sonet traffic.Including RPR on an Ethernet network is slightly different and may require a new network on top of the existing Ethernet. RPR rides on Ethernet's Layer 1 and can encapsulate Ethernet frames within an RPR frame. However, most Ethernet devices expect Ethernet to be the Layer 1 and the Layer 2 protocol of the frame, and won't carry RPR. Instead, the RPR equipment must be the core of the network, with Ethernet devices feeding in to or out of the RPR network.

Because RPR is a Layer 2 protocol, it defines a new MAC (Media Access Control) format into which data packets are placed (see chart "The RPR Header"). The RPR MAC header comprises 24 bytes. The first two bytes are for the ring control field, which is divided into seven subfields: TTL (time to live), P (parity), WE (wrap eligible), FT (frame type), FE (fairness eligible), RI (ring ID) and SC (service class). Service class determines the packet's priority on the network, while frame type determines whether the frame contains user data, fairness requests or control data for other nodes. Control frames contain node changes and other data. RPR supports network device and node autodiscovery; new network nodes announce themselves to their direct neighbors with control messages and distribute changes in their settings or topologies.

The next two fields are six bytes each and contain 48-bit destination and source addresses. These are hardware addresses as defined in section 5.2 of IEEE Std 802-1990. The Extended Ring Control takes up the next two bytes. The first eight bits contain the TTL Base. This is the original value of the TTL and is not decremented. The last byte contains the EF (extended frame), FF (flooding form), PS (past source), SO (strict order) and three reserved bits at the end, which should be 0s.

The EF field indicates whether the frame is a Base Data Frame or an Extended Data Frame. The Base Data Frame is used by all data traffic that travels from start to finish on the same ring. If data needs to travel from one ring to another to get to its destination, an Extended Data Frame, which includes the original source address and final destination address after the HEC (header error check), is used. The FF bits indicate whether data is flooded unidirectionally, bidirectionally or not at all. The PS bit is used during a wrap function to indicate the frame has traveled past the source address on its way back to the destination. SO is used when frames need to be kept in order.

The next two bytes contain the HEC, or header CRC (cyclical redundancy check). The protocol field takes up two bytes. When the value of the field is less than 1,535, the field indicates the length of the frame. If the value is equal to or greater than 1,536, it indicates the MAC client protocol. This value is designated by the IEEE Type Field Register. The protocol bytes determine type or length, never both. Finally, after the actual payload, a four-byte FCS (frame check sequence) is placed at the end.Like Sonet, RPR operates within rings. But RPR differs from Sonet in how traffic is placed onto the rings. In most Sonet setups, traffic flows in only one direction around the ring; RPR lets traffic be provisioned in both directions, thereby doubling the amount of traffic that can traverse the rings. Data is taken off the ring at its destination, which keeps the maximum amount of bandwidth between nodes available at all times. Some ring topologies, such as FDDI, keep data on the ring until it returns to the source, where it's finally removed. This would be a huge waste of bandwidth if the destination is nearby because the data would pass the destination and make its way all around the loop to be taken off the ring by the source node. (For a better understanding of Sonet and ring networks, see "Going Toward the Light," and "Sonet From Scratch.")Each RPR ring supports as many as 255 nodes. You can have multiple ring setups by placing nodes on two or more rings and requiring those nodes to route packets from one ring to the next.

Data traffic is placed on the resilient packet ring in one of three classes of service: high, medium or low. High-priority reserves network bandwidth that can't be used by any other traffic, even if the bandwidth is idle. Low-priority traffic can't reserve any bandwidth; it uses only unused bandwidth. Medium-priority traffic can reserve bandwidth on the network, but any unused portion is made available to other medium- or low-priority traffic. The medium setting is most appropriate for time-sensitive and bursty traffic. Bandwidth would be provisioned for the high points in the burst; during the low points, the unused bandwidth would be available to other traffic.

Although low-priority traffic can be provisioned to 100 percent of the ring bandwidth, high priority can occupy only up to 50 percent. High-priority traffic is guaranteed to be delivered even during a fiber break because RPR can use both rings concurrently. In contrast, Sonet typically uses only one ring and keeps the other for redundancy. If a break occurs, the high-priority traffic is routed to the other ring. Of course, enough bandwidth must be available, taking into account high-priority traffic that may be on the second ring.

Because low-priority traffic can occupy any unused portion of the network bandwidth, a node on the network could grab all unused bandwidth and leave nothing for the other nodes. To prevent this from happening, RPR implements a fairness routine that shares unused bandwidth between nodes.

During a fiber break or during normal operations, if medium- and low-priority traffic exceed what the node is capable of transmitting, the node will invoke Weighted Fairness, determined by a "weight" that is placed on each node. This weight specifies the priority that a particular node has in placing traffic onto the rings. Any RPR node can send fairness messages, but when congestion occurs at a node with high access to the ring, it will send a message to the other nodes on the ring to throttle back their traffic so that its traffic gets first shot at the bandwidth. When a fairness request is received, nodes sending data through the requesting node will queue or drop packets marked fairness eligible (FE) in their header. High-priority traffic is never considered for drop, so its FE value is ignored. Once a node is no longer congested, other nodes are allowed their share of available bandwidth.

Broken Ring

If a fiber break occurs, RPR offers two methods of getting traffic around the ring: Wrap and Steer. Cisco Systems and Corrigent supported Wrap, while Nortel Networks was a proponent of Steer technology. As a compromise, RPR supports both, though Steer is the default (see chart "Getting Around a Fiber Break").

Wrap is synonymous to Sonet's BLSR (bidirectional line switched ring) technology. Traffic is sent to the nodes on either side of the break, then to the opposite ring and to its destination. Wrap functions without much intelligence; traffic is simply rerouted around a break without considering its destination. Steer, however, provides intelligence at the source node. During a fiber break, the source node determines the best path to the destination and places the packets on the appropriate rings, disregarding the ring the traffic was provisioned for.

Wrap places lower requirements on the nodes but causes more traffic delays. This can be a problem for time-sensitive traffic, such as voice and video packets. Steer helps minimize delays, but it places higher CPU requirements on source nodes. Each source node must determine on which side of the break a destination lies and place the packet on the best ring. The priority of each packet and the available bandwidth also must be considered. Wrap and Steer function simultaneously on each node and are part of the provisioning process. Each connection created is specified as Wrap or Steer.

VariantsCorrigent is adding value to its products by combining RPR with MPLS. By provisioning connections end to end, the company is simplifying node-to-node processing and manual provisioning. And by supporting MPLS, Corrigent provides bandwidth management and segregates customers on metropolitan area networks. The addition of MPLS is an excellent way to provide end-to-end TDM (time-division multiplexing) service over RPR.

Because RPR is a work in progress, variations are being used by members of the RPR Alliance. For example, Luminous Networks markets its RPR as RPT (Resilient Packet Transport), a superset of RPR that includes full Stratum-level clock synchronization for TDM services. By bringing clock sync to Ethernet, TDM services like voice and video can be delivered with the same quality they would have on a Sonet network, without jitter.

Cisco also has a variant, DPT (Dynamic Packet Transport), that is supported on most of its Gigabit Ethernet and Sonet products. DPT was built around SRP (Spatial Reuse Protocol), allowing effective use of idle bandwidth. SRP is part of the final RPR specification.

All RPR vendors say they will support the final RPR specification when it's finalized in March.

Darrin Woods is a Network Computing contributing editor. Previously he was a Network Computing senior technology editor and a WAN engineer for a telecom carrier. Write to him at [email protected].