The slower time scales inherent to other types of national infrastructure have allowed development through careful advance planning. The explosive growth of the Internet and the shifting connectivity requirements of geographic communities present significant barriers to planned growth, however. New application communities entering the communications arena create sudden and swiftly climbing demands for bandwidth. Internet trading vastly increased connectivity requirements near cities with stock exchanges. The government focus on education infrastructure will lead to a similarly rapid but more sporadic growth near cities with schools.

In the absence of planning, federated ownership of communication networks has led to growth by aggregation. This approach does not mesh well with the infrastructure's inherently incremental expansion. New participants enter the field by purchasing a connection at the bottom of a vast hierarchy. As their connectivity needs grow, they must either reroute their connections to nodes higher in the hierarchy or pay to widen the nodes above them. The correct choice often changes as their needs continue to grow, periodically forcing widespread reorganizations to correct the sub-optimal incremental extensions. Resources employed to extend the network are often discarded during such reorganizations.

Growth of communication networks by aggregation suffers from two core problems, hazard expansion and geographic rigidity. A connection between end users in a strongly hierarchical, deeply imbedded network developed through aggregation typically employs many instances of line and concentrator hardware. Ownership of individual components along such a route is varied and typically unknown to the end user, thus each component is both a point of failure and a security hazard. As incremental aggregation lengthens the path between end users, these hazards continue to expand. Figure 2 below gives an example of the deeply imbedded systems that result from aggregation. A sample user at the University of Illinois may traverse a variety of infrastructures and service providers. In most cases, she may not select the service providers, resellers and vendors involved in her communications. Even if such choice were possible for the end user, it would be unacceptably difficult and onerous.

Geographic rigidity, the second core problem, refers to the inability of growth by aggregation to incrementally adapt an existing backbone network to changing geographic requirements. Owing to unpredictable growth patterns and the expense of reorganization, the geographic locations of switches in a backbone are rarely optimal, and the point of closest physical access is typically along one of the network links rather than through a switch. New connections are generally routed through an existing switch, however, requiring longer fibers and increasing the load on concentrator hardware at the switch. When the new connection grows to the limit of the switch or concentrator, the backbone is reorganized to incorporate a new switch, and the suboptimal lines are discarded.

Our program investigates robustness and recovery in direct access networks, an innovative approach to the incremental construction of reliable, scalable, and affordable high-speed networks. Figure 3 shows the architecture we propose. Direct access networks address the problems of growth by aggregation in a cost-efficient manner. Land-line connections provide access to land-based aggregated traffic for conventional access, e.g. via telephone lines, and afford access to wireless users, who may use different types of services and applications. We consider access networks where the access land-line network is a high-speed optical network. The land-line network in the upper left-hand corner represents conventional use of the land-line network. Note that aggregated wireless users, for instance conventional mobile telephony users, may access the network in this manner. The second type of users are wireless users who access the network via wireless access nodes, e.g. base stations, connected to the land-line network through access ports. Such direct access provides three significant advantages. First, it gives users access to a variety of services, e.g. different rates for data transfer. Second, direct access uses the existing, geographically dense wireline infrastructure. Finally, direct access allows for simple extensibility and flexibility of the access network, for instance by changing the types of access nodes at access ports. The user on the upper-right hand corner of figure 3.a represents a wireless user connecting to the access network via wireless links represented by dashed lines. Such a user may use conventional wireless-telephony services, or run nomadic computing applications requiring more bandwidth and entailing different quality of service (QoS) requirements than wireless telephony. The user may gain access to the wireline network through one or more access nodes. Figure 3.b shows two data streams, emanating from two access ports connected to access nodes. These two data streams may both correspond to data transmitted by the user depicted in Figure 3.a. Both data streams are sent to some common destination, which may be the final destination or some intermediate destination, such as a processing node which processes the two streams to yield a single stream from the wireless user.

Direct access networks offer significant advantages over traditional aggregated networks in terms of extensibility, both for permanent use and for dynamic utilization. Access ports may be added, within bandwidth and network management constraints, for more graceful growth than the aggregated model allows. First, as illustrated in Figures 2 and 3, less wireline infrastructure may need to be deployed to introduce a new access node. Second, a single switching/routing node may not need to shoulder the full burden of a new access node. For instance, in Figure 3, the traffic to and from Access Node 1 at Access Port 1 may be evenly distributed between the two land-line switches shown in Figure 3.a. Another example is the case where a great part of the traffic is local among adjacent access nodes and does not require the intervention of a switching/routing node. Such an example would occur if Access Node 1 at Access Port 1 receives wireless signals for retransmission by Access Node 2 at Access Port 2. The high-speed network is then an aid for retransmission traffic emanating from Access Port 1 is terminated at Access Port 2. If the two access nodes had instead been connected to two switching nodes, those switching nodes would have had to handle the full traffic between the access nodes.

Dynamic extensibility relies on having access ports may be in use or idle, depending on current needs. For instance, access ports could be leased for a week during special events such as conferences or Olympic events. An access port may be used by a computer network offering Internet access to conference attendees or video feeds from an event. In case of catastrophic events, access ports may be used to accommodate wireless links to replace destroyed wireline connections. The ability to lease such an access port rather than purchase a permanent installed base allows capital expenditure to be amortized over several different types of users, many of whom may be only occasional users. Access nodes which may be set up for use in a few hours or less can therefore offer flexible, on-demand direct access for a variety of access services.

In light of our previous discussion, we may define the action of the different types of nodes and ports. Access nodes connect to access ports and interact with the access network through the ports only. Access ports, while ultimately controlled by the access network, respond to requests from access nodes and manage all communications to and from the access node. The access network has full control over links and routing/switching nodes. The access network also determines the range of services available to access nodes through their attached access ports. The access ports are thus under joint control of the access nodes and the access network, but the latter retains ultimate control.

Access nodes could more accurately be described as limited access nodes. Indeed, access nodes use only a particular subset of possible services. The access ports to which they connect provide, in turn, access to a limited portion of the bandwidth and the services available from the backbone. Note that, depending on the type of access nodes and their needs, access ports may access different services, levels of service or bandwidths at different times. There are three main reasons to keep access nodes and the operation of their corresponding access ports limited.