Packet rings aim at metro nets, part 2
Deterministic performance
Performance is another important characteristic of the access network. For convergence over IP to become a reality, the access network must be capable of supporting services with deterministic and predictable performance. For example, real-time packet voice and video applications require minimal latency and jitter. Applications such as data vaulting that are intolerant of delay and data loss must be supported by connections offering guaranteed bandwidth, low loss and minimal packet delay. The ability to ensure performance guarantees by traffic type will enable network operators to support integrated services on a common access network.
Last, the access network must be efficient. It should be optimized for the ring topology of the MAN fiber plant and for the predominant traffic type it carries-data. Circuit networks and technologies are not readily adapted to data traffic and are often compromised when data capability is introduced. In a data world, a packet network is the logical infrastructure. The packet access network must incorporate service protection. However, because of the high cost of the MAN's fiber-optic infrastructure, utilization of that fiber must be efficient. Therefore, service protection mechanisms must be far more efficient than the 100 percent sparing used in Sonet. In addition, the network must apply traffic engineering concepts to manage and balance traffic efficiently and deterministically.
The RPR standards will define a MAC with two network interfaces, a system-level interface and a physical-layer interface. The topology associated with RPR will be a set of switching nodes interconnected along a bidirectional ring.
To connect adjacent nodes on this ring a variety of media can be used, including a pair of dark fiber strands, a pair of wavelength-division-multiplexing-derived wavelengths, or a Sonet add-drop OC-n circuit or other bidirectional connection medium. Based on the topology, each RPR node supports two ring ports; one supports a connection to the adjacent node to the left, the other to the right.
The MAC interface for RPR fundamentally differs from an Ethernet MAC in that it contains elements of a switching protocol, including decision making for storing and forwarding data. This key difference of bringing switching decisions into the MAC provides significant network performance improvements compared with Ethernet protocols.
Just how RPR will address the MAN requirements is reflected in the early thinking going into the RPR standard. Since the standards effort is just beginning, these ideas are expected to change.
Resiliency: The physical layer will detect faults and signal that information to the MAC layer. If the failure is determined to be critical, each affected RPR node will initiate a fail-over action for the service flows it originates that are affected by the facility outage. The fail-over action is a simple redirection of the traffic from the failed path to the protection path. The process of alarm notification and redirecting traffic will be completed within 50 ms of the outage.
Fairness: RPR will use a combination of bandwidth reservation and quality of service to manage bandwidth allocation on the ring. Traffic is classified, policed and buffered on entering the ring. Packets admitted onto the ring shall be switched through intermediate nodes with only nominal delay. The destination node removes the packet from the ring and performs any additional processing required. The RPR algorithms will coordinate ring operation, ring access and traffic management among all the nodes.
Scalability: RPR supports a number of physical media and line rates with initial proposals, including line rates up to 10 Gbits/s. Besides the raw bandwidth capacity of the ring, RPR will incorporate labeling schemes that will give the operator the ability to manage millions of service flows on a single ring. This combination of millions of service flows and superbroadband ring aggregation is exactly what is needed to support the burgeoning data demands on today's access networks.
Efficiency: RPR systems are packet switched, so that by their very nature they use bandwidth more efficiently than circuit-switched systems. RPR uses dual counter-rotating optical rings, with data being transmitted and load-balanced on both rings. Destination stripping is applied to conserve ring bandwidth. Unicast packets are removed from the ring by the destination RPR node. Moreover, RPR packets do not circulate the full ring as is required by ring-based LAN technologies. This design can multiply the effective bandwidth of the ring several times depending on traffic distribution around it. The combination of packet switching and spatial reuse creates very efficient ring networks, optimizing the utilization of the fiber links connecting the RPR nodes.
Performance: Finally, an important construct is the RPR service flow or virtual media, defined as an edge-to-edge element, from ring entry to ring exit. The service flow includes a label that provides RPR with a handle for quickly processing traffic transported around the ring. An RPR node need only inspect this label and one or two other header fields to determine how the frame should be processed. Frames that are not destined for the node are shunted through with minimal delay and jitter. Frames addressed to the node are removed from the ring and sent to one of the node's service modules for additional processing.
This cut-through capability of RPR, combined with the RPR fairness algorithms, will enable vendors to develop RPR systems that ensure both bandwidth and delay performance per application. Indeed, we anticipate that this degree of control and service assurance will exceed the capabilities of ATM in the access network.
The IEEE 802.17 RPR standard will be developed on a very aggressive schedule, with its technical requirements established in the next 18 months. To meet this timetable, a number of vendors have banded together and formed a Resilient Packet Ring Alliance, to work cooperatively with the 802.17 Working Group and to sponsor early interoperability testing. Reuse of standardized physical layers will remove a major inhibitor to rapid standards development, allowing the Working Group to focus on the MAC specification. |
