Reliability in optical networks is usually achieved in hardware, either through active duplication or dynamic restoration via shared backup fibers. The algorithms used for these purposes are typically designed to strike a particular balance between restoration speed, capacity efficiency, and failure resilience. Recently, we introduced an approach [1,2,3] that provides superior robustness to failures yet allows network management to dynamically tune between available capacity and network reliability. As optical switches remain extremely expensive, we have also developed protocols that extend backbone reliability to inexpensive access ports [4], an approach made practical by recent successes in optical header recognition [5,6,7]. Together, this work forms the basis for the definition of optical domains, which extend the reliability traditionally enjoyed by backbone networks down to end users. Optical domains utilize one or more wavelengths within the domain to provide a shared medium for low-bandwidth communications. They simplify the interface between the WDM layer and routing protocols by supporting hysteresis in lightpath allocation: small amounts of point-to-point traffic are routed through the shared wavelengths; heavier flows receive private lightpaths.

High-capacity optical systems typically use WDM. Accessing only full wavelengths, however, is very restrictive. Low-bandwidth flows are thus typically aggregated in a hierarchical fashion before entering an optical backbone at a switch. To accommodate such flows, packetized, TDM, and CDMA systems have been developed. CDMA systems suffer from the difficulties of processing at optical rates. TDM systems suffer from timing jitter over long spans. Packetized systems allow the users to access the system without the timing considerations of TDM and do not require the processing of CDMA.

Regardless of the choice of protocol for sharing wavelengths, the common use of non-redundant aggregation units prevents the level of reliability built into the backbone network from extending to users. Supported by a next-generation Internet (NGI) grant from DARPA, we are developing protocols to support the extension of all-optical networks with inexpensive access ports that address wavelength sharing in a way that provides end users with backbone-level reliability. Unlike optical switches, access ports perform no routing or aggregation; rather, they allow injection of a relatively small amount of traffic and recovery of a small amount of traffic from a set of fibers. For instance, an access port may only be able to access a single wavelength in a WDM system, or one set of time slots in a TDM system, or only read packets with a certain header in a packetized system. An example application of access ports is that of wireless base stations distributed roughly along the path of a backbone and connecting to the backbone via access ports. Another example of access ports is gateways for computer networks connected to the backbone.

Logically, the presence of access ports in a network decouples access from routing, enabling more cost-effective approaches to network growth. Access ports afford flexibility, by allowing points of access to be added or removed without affecting the basic topology of the network. Moreover, access ports do not possess the hardware of switching nodes and are far less costly. Reliability in such access networks depends on the reliability measures in wireline backbone networks, on reliability in the peripheral systems attached to the backbone networks via access nodes, and on the interplay among access nodes, the backbone networks and the peripheral systems.

Our research has focused on localization of failures in access networks and on the impact of backbone network protection and restoration on access nodes and peripheral systems. As a part of this research, we have also developed insight into the relationship between backbone network capacity and reliability [2,3].