Conventional transport systems rely on electrical switching to reroute traffic around facilities failures. Many restoration architectures have been deployed and proposed, including 1:1 and 1:N linear protection schemes, SONET/SDH ring protection schemes, and a variety of shared capacity mesh restoration schemes. These schemes vary in cost and performance, but in all cases, additional electrical network capacity is allocated for connections affected by the failure to be rerouted electrically to an alternative path through the network while the fault is corrected.
An agile network that allows fast connection setup is appropriate for such a recovery function—"redialing" affected wavelength connections over backup facilities when network faults occur. We define an agile photonic network as a network of fibre-optic links interconnected with photonic switches (without optical-electrical-optical conversion), equipped with broadly tunable source and filter structures at the network edge, and able to set up an end-to-end wavelength path without manual intervention (see Figure 1).
Many architectural concepts employed in conventional electrically switched networks may be applied at the photonic layer (e.g., shared protection rings, shared mesh restoration, and 1:1 path switching).
Model 1: Client layer 1+1 switching. As a baseline architecture for comparison purposes, we use a network model where an electrical protection switch at the edge of the network provides rerouting onto the dedicated redundant path (see Figure 2). In this case, the photonic layer may provide agile wavelength setup for provisioning, but does not react in case of a fault. The edge electrical switch in practice may be integrated onto the transponders or in a subtending service platform such as a TDM-based STS crossconnect or an IP router. This architecture provides significant cost savings compared to conventional electronic-crossconnect (EXC)-based mesh networks using point-to-point WDM systems.
Model 2: Photonic 1:1. Redundant facilities are also dedicated through the network for restoration, but the connection is switched in the photonic layer using the edge tuning and photonic switch rather than an electrical switch. Here, the protection wavelength connection is set up across the network beforehand but remains "unlit" (since the transponder is currently lighting the working path) until the fault occurs. Crucially, there is then no redundant transponder, rather the edge transponder is a key element in the switching event, operating in conjunction with the photonic switch and flexible multiplexer/demultiplexer at the path end points to light the protection connection. That means the system can recover from all line faults, but the end-point transponders remain a single point of failure in the system, unlike in model 1, where transponders are duplicated for the protection path.
Model 3: Photonic shared mesh. Instead of allocating dedicated protection path facilities for each connection, a shared pool of network resources, including wavelengths and regeneration facilities as required, is allocated for network restoration. When a fault occurs, affected connections are reestablished across unaffected paths that utilise the shared resources. This approach requires that the restoration resources be dimensioned sufficiently to ensure all affected connections can be rerouted for every failure scenario. It is further complicated in an all-optical network by the need to ensure that restoration path regeneration be placed as required for very long paths. Like model 2, the ingress and egress transponders are inherently single points of failure in this restoration scheme, implying an inherently lower reliability than model 1.
Analysis (in our model) is based on a 50-node backbone network covering the major population centers in North America. Optical connections between sites are at 10-Gbit/sec granularity; the traffic model is a 3-Tbit, 110-node gravity model consolidated through hubbing into the 50 major nodes sites on the backbone. More weight is placed on population than distance between cities to reflect emerging demand patterns. For each of the three above-mentioned network equipment models, the network is configured and transport costs are calculated and illustrated in Figure 3.
An unprotected network model provides a reference point. Line equipment costs include the amplifier and photonic-switch costs. Models 2 and 3 show a significant network cost savings through elimination of edge transponders (since they are no longer duplicated) over model 1. The cost of the electrical-switch ports have not been included in this comparison but could add a significant extra cost to model 1.
Model 3 also shows additional line and regeneration cost savings due to sharing efficiency of the mesh scheme. It is notable that the network-cost premium for the mesh network is not dramatic compared to the unprotected case. This observation is due to the fact that in an agile photonic network, cost is dominated by the edge transponders, and therefore the additional line and regeneration cost to support photonic layer restoration does not represent a significant premium over the unprotected case.
This network cost analysis shows significant cost variance, but these network models do not deliver equivalent connection reliability performance. Model 1 provides hitless (sub-50-msec) edge switching and fully redundant transport equipment. Models 2 and 3 only provide line redundancy, and 50-msec switching is likely to be difficult to achieve in the near-to-medium-term. To evaluate these differences, we can compare the models based on two metrics: connection downtime and number of service "hits."
For this comparison, we consider a protected reference connection through a network whose working and protection paths are identical with the following typical long-haul attributes:
- Distance: 1,500 km.
- Fibre-cut frequency: one cut per 1,000 km per year.
- Equipment failure rates: line equipment—one failure per 1,000 km of system length per year; end-point transponder and multiplexer/demultiplexer—25-year meantime between failures; meantime to repair—fibre cuts eight hours and equipment failures four hours.
A connection redial time of 10 sec is also assumed. We used these assumptions in a system reliability model to estimate the expected downtime and number of service hits on the connection for each network model. The following Table summarises the results, including an unprotected model for comparison.
For these metrics, Model 1 provides the best reliability performance. The single point of failure at the transponders and the redial time introduces an average of more than 12 minutes per year downtime to the connection. Use of a 1:N electrically switched transponder protection scheme for transponder failures coupled with photonic layer restoration around line failures would dramatically improve connection availability by eliminating the transponder single point of failure.
Such hybrid restoration architectures are certainly worthy of further study. Also, the number of service disruptions due to redial events is high. These specific results will vary widely according to network and equipment specifics, but the basic performance versus cost tradeoffs are expected to remain consistent.
Examining the cost and performance of using photonic layer redial for network fault recovery is instructive. While significant end-point transponder costs can be eliminated from the network by using photonic-network-based restoration schemes, these savings do not come without a performance penalty in terms of connection availability and the number of customer-visible connection interruptions experienced. There may be a number of applications for this approach.
For IP switch trunk carriage, emerging, fast, packet-based recovery schemes can be employed for high-priority traffic, while photonic layer recovery may be sufficient for lower-priority traffic. As an enhancement over current unprotected wavelength services, photonic restoration may offer a significant performance enhancement at acceptable additional cost. Used as a "second line of defence" for current premium service protection schemes to further enhance availability, agile network redial may also be applicable. The potential value of these applications depends on the specific requirements of the underlying services carried and subtending service layer network architecture.
Peter Roorda and Greg Friesen of Innovance Networks (Ottawa, ON) can be reached at proorda@Innovance.com.