Once installed in a service provider's network, basic infrastructure equipment is likely to remain for a very long time. As customer needs evolve, carriers periodically add new features and services to existing systems instead of replacing those systems with new technology. As an example, the T1 circuits that connect the building where we work to our service providers are here to stay. We can lease more T1s (we are too far from the central office for a T3), but there is no economic justification for new equipment or fiber deployments.
Ethernet is inherently distance-limited, so if my company would like a high-bandwidth Ethernet connection to our Internet service provider or between two of our locations, how can our carrier provide it? A practical solution is circuit-bonding. Bonding creates a single high-bandwidth connection between equipment from a set of smaller connections. Circuit-bonding also adapts the payload to the underlying connections. In my Ethernet example, several standard T1s can be bonded to support the Ethernet connection.
While adapting and transporting new signals over an existing network is an important bonding application, creating a big pipe has other advantages—especially if the bonding equipment enables the network to support multiple payloads. Several bonding techniques exist, but flexible bonding systems that are capable of supporting any type of payload, using any type of transport channel, have only recently become available. A newly developed "universal transport protocol" (UTP) provides the foundation for these systems.
Nearly every service and transport platform encounters situations where it is beneficial to support some bonding features. The need has been recognized repeatedly, and the results are protocols like inverse multiplexing over ATM, multilink point-to-point protocol, multilink Frame Relay, load sharing in routers, and SONET virtual concatenation. Each of these protocols is specific in its underlying protocol or format, and each produces a unique result.
These protocols work well for their particular applications. The drawback is that each technology performs differently, meaning a service provider cannot offer consistent services across platforms. These technologies were not designed specifically to solve bonding issues, but they are beneficial additions to platforms already deployed.
None of these protocols is particularly well suited for the application of providing our company with a wide-area Ethernet connection, however. This application and many others require a new technology that addresses issues of transporting any payload over any imbedded service-provider infrastructure. The UTP refines the application of bonding to answer this need.
Bonding has been described as combining the capacity of channels in parallel. However, several additional features are required to support operation in a service-provider network where long distances are common and reliability and OAM&P (operations, administration, maintenance, and provisioning) are critical. The UTP can be implemented in a multiblock hardware architecture to provide these capabilities.Since the amount of bandwidth available to the system, for example, is limited and subject to change due to channel failures, payloads entering the system must be regulated. The UTP multiblock architecture accomplishes that via the rate control block shown in Figure 1. This block limits data clients and ensures TDM fixed-rate clients have the appropriate bandwidth. Next, the shared access block provides the intelligence to determine which payloads are awarded bandwidth at any given time based on provisioning parameters and operating conditions. These functions are critical to support multiple payloads of different types and different priority levels in a single system.
The bandwidth concatenation block is responsible for mapping the payload data into the operational channels. That is perhaps the most obvious function in the bonding system. For a system supporting multiple payloads, efficiently using available bandwidth, and assigning priority levels to payloads, this function is nontrivial.
Another basic function is to respond to failures and impairments within the individual channels. The protection blocks provide this function. Since payloads are dispersed across the individual links, a failure in any link causes a disruption. Therefore, each of the channels in the link must be monitored closely, and the appropriate response to failures must be made quickly. The response may result in the (1) continued full support of all payloads if enough capacity remains in the link, (2) reduction of capacity available to data clients, or (3) removal of the entire payload in the case of a TDM client requiring more capacity than is available.
Figure 1 shows the protection blocks responding to a single failed link. In this case, the protection blocks at the ends of the link have coordinated the removal of the failed channel, and the payloads use the remaining operational links. When the failed channel resumes normal operation, the protection blocks convey the availability of the additional channel and add it automatically.
Another issue is the effect of transmission paths of different lengths. Each of the channels composing the aggregated link may travel a different path and experience a different propagation delay getting to the receiver. The result is a differential delay among the channels that must be compensated by realigning the channels at the receiver. The compensation is provided by the skew compensation block.
The payload extraction block serves to reconstitute the payload signals from the aggregated link. It uses information provided by the shared access block. Lastly, the reconstituted payload is delivered to the appropriate port.
A transport system incorporating the described bonding functions can efficiently and economically support a variety of applications. One example is an Ethernet private line service. Figure 2 illustrates the transport of Ethernet using embedded transport infrastructure. The bonding system allows Ethernet private line connections between enterprise routers and Internet service providers. In this scenario, the enterprise can enjoy the full flexibility of Ethernet while the service provider capitalizes on existing infrastructure.
The full benefit of bonding is realized when bandwidth aggregation, normally associated with the term "bonding," is coupled with bandwidth-sharing among multiple payloads and with protection capabilities. Operating in concert with existing transport systems, these functions extend the capabilities of transport networks. Although there are several higher-layer protocols with bonding features, bonding is most efficient as a simple low-layer transport function that enables any payload to be supported over any transport network.
As enterprise bandwidth requirements increase independent of transport rates, bonding is a practical solution to provide the necessary bandwidth. Bonding serves to sever the unwanted link between transport technology and payload bandwidth requirements, allowing the support of a variety of services over unlimited distances with efficient and economical transport systems. With the appropriate bonding and adaptation technology in place, service providers are not forced to upgrade transmission equipment to support the latest payload protocols or data rates. Therefore, transport platforms can evolve at their own technological and economical pace, while continuously supporting the latest services.
One benefit of deploying a comprehensive bonding technology that is gaining attention is the ability to offer Ethernet private line services using existing transport infrastructure. Service providers are currently using this option to offer new services with their existing infrastructure. In the future, as new products, rates, and services are offered, the flexibility bonding enables will continue to allow easy integration and support.
Dr. Robert K. Butler is principal engineer and co-founder of Ceterus Networks (Allen, TX).