Star versus ring topologies in metropolitan-network applications

Th 09spr06 1

Metropolitan access rings use Synchronous Optical Network/Synchronous Digital Hierarchy (SONET/SDH) add/drop multiplexers (ADMs) to transport voice and data from remote points of presence to the service provider's central office. Such ring configurations allow for protection from fiber cuts and performance monitoring. The advent of intelligent dense wavelength-division multiplexing (DWDM) solutions for the metro, however, has enabled the development of alternative architectures. It's now possible to map wavelength connections onto a physical ring of fiber so that all nodes are connected logically in a star topology. These systems can offer the same protection and monitoring characteristics currently available from SONET/SDH nodes, essentially over the same fiber rings. Given that these features can be incorporated either directly into router ports, or into the DWDM transport equipment, the question arises as to which is the more efficient topology for router interconnection.

Ring-based systems such as SONET/SDH ADMs offer a division of the given line-rate bandwidth among users connected on the ring. An OC-48 (2.5-Gbit/sec) ring, for instance, can divide into four OC-12 (622-Mbit/sec) ports, 16 OC-3 (155-Mbit/sec) ports, or some combination of the two. But this rigid division of bandwidth is inefficient for packet-based traffic. Why limit an OC-3 port on an OC-48 ring to just its own portion of the bandwidth if none of the other users on the ring are fully utilizing their portion?

Several vendors have addressed this problem by offering statistical multiplexing of data onto SONET/SDH lines. One example is Cisco Systems' Dynamic Packet Transport (DPT) in which SONET/SDH line protection and monitoring, as well as statistical multiplexing, are incorporated into router ports. With this protocol, there is an additional advantage of re-using spatial portions of the ring. So if a packet progresses halfway around a ring on average, there is effectively twice as much bandwidth shared among the users. Such a protocol also allows for the use of both active and protective pathways on the ring with the total bandwidth dropping to the standard value during a fiber cut. It appears that bandwidth is efficiently partitioned by connecting routers in a ring-type configuration.

The same features can be incorporated into the intelligent optical layer of DWDM equipment. To protect against a fiber cut, two different pathways are used for each connection: one clockwise and the other counterclockwise on a ring. If one pathway is cut, there is an alternate route for the connection. Intelligent DWDM systems, such as the New Access MetroFusion system, can then multiplex data from the blocked pathway to the unblocked one. Just as in the case of DPT, the total bandwidth is doubled without a fiber cut and it drops to the standard value if a breakage occurs.

The re-routing of data around a fiber cut must be accomplished within the standard of 50-msec set by SONET/SDH. With equal recovery and monitoring features, how then does a star-based system compare with a ring-based system? In the case of a ring where the line rate is shared among all the users of the ring, the throughput per node is twice the line rate, divided by the number of nodes on the ring multiplied by the bandwidth re-use factor. For analysis, assume that, on average, a packet travels halfway around the ring, giving a bandwidth re-use factor of 2 and a throughput per node of 2 x 2B/N, where B is the line rate and N is the number of nodes. Th 09spr06 1

The double-ring and star configurations have an equal cost basis while single-ring is half that cost.

Key to this comparison of the different topologies is the cost associated with a given throughput. The characteristic parameter for this analysis is the number of router ports used for interconnection. If DWDM transceivers are used, the router port count is equal to the number of transceivers. If non-DWDM techniques are used, the number of router ports equals the number of fiber spans.

Internet service providers (ISPs) that use the transport services of a competitive local-exchange carrier (CLEC) or a regional Bell operating company (RBOC), lease a pipeway with a given bandwidth. These ISPs do not care how the transport layer is implemented--whether it is through the use of more fiber, SONET/SDH multiplexing, or a DWDM system. Their major concern is the costs associated with the number of router ports they must use and with the pipeways they must lease.

Using the number of router ports to determine the associated costs provides a practical, vendor-independent analysis of the expense of a given configuration. For example, each node in a logical-ring topology requires a router port for both the clockwise and counter-clockwise directions. The throughput from a single ring may not always provide enough bandwidth for a given application. To increase the throughput bandwidth, another ring can be added in parallel. Thus, in this example, four router ports are needed at each node to double the throughput of a single ring. A comparison of the throughput bandwidth will determine the given cost in a ring topology versus a star topology.

The star, or hub, configuration is a very familiar topology in networks. It is not commonly used in metropolitan area networks (MANs) because of the high fiber count associated with the architecture. Yet, with the advent of DWDM, such a configuration can map onto a physical ring of fiber. The transmission equipment provides fiber-cut protection as well as performance monitoring, which is absent in a traditional star network.

In the case of a dual-homed star network over a ring, the number of router ports required per node is easily calculated: two at the hub and two at the edge. The throughput per node under normal operation is simply twice the line rate. With this information, network planners can determine which architecture is the more cost-effective solution for a given number of nodes. Using the router port count required by each configuration to determine cost, a double-ring has an equal cost basis to that of the dual-homed star. The single-ring configuration is half of that cost. The bandwidth throughput, however, is much smaller for both ring structures in contrast to the star configuration (see Figure).

Therefore, if a single-ring topology provides enough bandwidth for a given application, then that configuration is the preferred choice. If, however, higher bandwidth is required, it is always more advantageous to deploy a star versus a double-ring structure. The star topology offers a lot more bandwidth at an equal cost to the double-ring network when four or more nodes are present on the ring. For example, given an OC-48 line rate and 16 nodes, the average throughput for a single-ring is OC-12, it's OC-24 (1.25 Gbits/sec) for a double-ring, and it's 2 x OC-48 (5 Gbits/sec) for a dual-home star. A simple calculation also yields that there is an N/4 (where N is the number of nodes) reduction in port count using the star topology. Therefore, in a 16-node network, it is possible to reduce the associated costs by a factor of four.

The star topology that uses intelligent metro DWDM is a viable, cost-effective alternative to shared-ring configurations. In cases where high bandwidth throughput is required from each node, the star network provides significant savings over shared-ring systems.

Traditionally, only ring-topology systems such as SONET/SDH provided the protection and monitoring necessary for building reliable networks. Today, next-generation intelligent DWDM systems provide these features without restrictions on topology.

Near Margalit is president and chief executive of New Access Communications (Santa Barbara, CA).

More in Network Design