Making network survivability more affordable, ATM enables service providers to back up their networks without fully redundant facilities

As businesses around the world increasingly rely on telecommunications to gain a competitive market edge, intelligent networks globally interconnecting businesses and financial communities are growing in size and number, as well as in the number of new applications.

Price, which relates to network costs, and reliability, which relates to network survivability, are business customers' predominant criteria in choosing a telecom service provider. Although price has been the primary concern, reliability is rapidly gaining more importance.

With high-capacity fiber optic transmission systems forming the backbones of most current networks, survivability of traffic against network failures has become ever more critical. Many of the emerging business telecom services, particularly the fast packet asynchronous transfer mode/frame relay-based applications, require more robust transmission systems. Network outages and performance degradation have a much more severe effect on these applications than on conventional services.

Conventional network survivability Transport network survivability is usually measured by its long-term availability or average network up-time. Most businesses expect their communications links to be continuously available to minimize potential loss of revenue. High network availability at the transport layer may be achieved using millisecond restoration schemes provided by self-healing network configurations such as fast facility protection (FFP) or Sonet rings, both of which are used in Stentor's Canadian network.

An FFP network comprises two physically diverse routes with identical transmission systems (Figure 1). Each route carries half of the working traffic and half of the restoration traffic. The restoration traffic on each route is the duplicate of the working traffic on the other route.

To maintain high reliability over the long haul, the network should be subdivided into smaller sections. This may be accomplished using crosslinks that connect the two routes together, much like ladder rungs. Digital cross-connect systems (DCSs) located at connection nodes may then be used to dual feed and receive individual facilities-for example, DS-3s-within shorter sub-sections. An interruption of signal reception at a DCS port due to a facility failure on the primary route would cause it to switch to its hot standby port to receive restoration traffic, typically within 50 milliseconds. This ladder-like backbone improves network availability because each section can be protected against a single route failure, reducing the potential for outages due to simultaneous failures on both routes.

A series of interconnected Sonet rings can also provide a self-healing network architecture (Figure 2). The dual route structure of the rings allows duplicate signals to be sent in opposite directions. In the event of a failure in a ring, the network switches to the route carrying the duplicate signal in the opposite direction-away from the failure point. Because the protection switching action typically would occur within 50 milliseconds, customers would not notice that a failure had occurred in the network.

Inefficiencies These self-healing schemes use fully duplicated transmission systems and capacity for alternate routing of today's time division multiplexed (TDM)-or synchronous transfer mode (STM)-circuit facilities.

The restoration capacity and the associated transmission systems are essentially unused, except in the rare occasions of network failure. Furthermore, the STM transport facilities suffer from inefficiencies inherent in capacity usage-for example, low-fill factors-due to TDM's rigid multiplexing hierarchy and the use of a limited number of discrete transmission rates.

Also, the majority of high-growth business applications such as local area network interconnection are bursty and asymmetrical. These applications fit poorly into STM's rigid multiplexing hierarchy, making inefficient use of network resources. When capacity is dedicated for restoration purposes-as in a self-healing configuration-inefficiencies are doubled.

This inefficient use of bandwidth and transmission facilities makes survivability very costly for today's networks. As a result, some carriers may hesitate to provide network survivability on an end-to-end basis.

A partial solution can be found in the "service layer" protection offered by dynamic control routing systems associated with circuit-switched telephony. Although mainly intended for remedying congestion on voice trunks, a dynamic control routing mechanism could also be used to provide restoration of switched voice traffic in network failures. This could be achieved without having to dedicate complete restoration capacity by sufficiently increasing the overall number of trunks on alternate routes to meet certain performance objectives.

The performance or survivability at this service layer is usually measured in grade-of-service metrics such as the percentage of calls blocked and dropped. This solution only works for switched telephony traffic, however.

Another method of survivability used by some carriers to avoid dedicating full restoration bandwidth is at the "logical layer," based on the application of dynamic routing schemes in conjunction with DCSs. In the event of a failure, a search would ensue to find alternate routes with sufficient idle capacity to temporarily handle the traffic of the failed facilities. However, the current recovery time of these DCS-based mechanisms is long-approaching several minutes.

ATM's Role To date, many telecom service providers have concentrated on finding service-specific killer applications for ATM. However, no clear candidate has been identified yet. As a result, the economic justification for bringing ATM into commercial existence is still unclear.

For ATM to succeed, carriers should consider shifting their initial focus toward technology-based applications that could improve network efficiencies and cost effectiveness. Because the voice telephony network is highly ubiquitous and mature, the first ATM applications will most likely be in data communications networking. In this scenario, ATM would first be implemented as an overlay core network, catering to data and private line services.

One area with high potential for network cost reduction is network survivability. ATM's statistical multiplexing and non-hierarchical path structure, as well as its cell-based dynamic routing and traffic management, may be exploited to simplify survivable network architectures and make much more efficient use of the available capacity.

Statistical cell multiplexing is a key ATM capability that allows different users and services to share each other's spare capacity. Unused bandwidth does not have to go to waste but can be reclaimed by other users and services. Because peak times for different users and services do not coincide, multiple users and services can be consolidated.

ATM categories To meet the bandwidth delivery requirements of various applications, several ATM service categories have been defined. Each category has a different set of quality of service (QOS) traffic parameters. Among the five categories defined to date, constant bit rate (CBR) is the least flexible and available bit rate (ABR) is the most flexible in terms of transmission efficiencies.

With CBR, delivery of a fixed cell rate-referred to as peak cell rate (PCR)-is guaranteed through the ATM network. CBR has been defined primarily to accommodate time-sensitive circuit-switched applications-for example, circuit emulation of switched voice telephony. CBR can waste bandwidth because of the additional overhead of ATM packetization-another reason for keeping the existing switched telephony network separate from an ATM overlay.

The variable bit rate (VBR) categories-VBR real-time and VBR non real-time-also have a PCR that cannot be exceeded, but the guaranteed bandwidth is a lower sustainable cell rate. The direct output of a video codec is an example of a suitable VBR application. Both CBR and VBR require policing for their peak bandwidth usage.

The traffic management schemes of ABR and unspecified bit rate (UBR) are more suitable for business data communications, especially the bursty traffic of local area network interconnections. UBR and ABR both use available bandwidth for transmission after capacity requirements of other service categories have been met. However, while there is no guarantee of bandwidth availability and no flow control mechanism using a UBR traffic contract, the ABR category can offer both.

ABR offers a minimum cell rate option and several flow control mechanisms to provide some level of guaranteed bandwidth and data transmission integrity. A PCR has also been defined for ABR service, which can be set at the maximum information rate of the end user terminal interface. ABR is a very promising category, especially for data communications applications. The virtual connections belonging to the ABR and UBR QOS categories can be maintained without wasting resources.

Virtual circuits and virtual paths A number of advantages result from the flexibility of ATM's virtual connections and QOS options. The virtual characteristic of ATM connections-including virtual paths and virtual circuits-allow automatic traffic rerouting by simply changing the values of the virtual path or virtual circuit identifiers in cell headers. With appropriate signaling systems, fast packet ATM switches or cross-connect systems could provide dynamic rerouting and call re-establishment.

For example, using something similar to the service layer restoration capability of dynamic control routing systems in switched telephony, ATM networks will be able to support restoration of switched virtual calls, or connections. Key to this application are end-to-end virtual circuits, established in real time on a call-by-call basis using ATM's Q.2931 signaling.

However, because ATM uses a common packetized format for transporting all types of services, this dynamic reconfiguration capability is not service-specific and could apply to all types of voice, data, image and video applications.

As with circuit-switched telephony, the use of dynamic control routing for ATM's switched virtual circuits (SVCs), would eliminate the need for dedicated full restoration capacity. Instead, depending on service and traffic patterns, some extra capacity on alternate routes would be needed. ATM's statistical multiplexing will help to optimize this extra capacity. A carrier might use SVC-based dynamic control routing for data and video traffic restoration. Currently, however, networks do not support Q.2931.

As an alternative to virtual circuit-based restoration schemes, a number of virtual path-based self-healing schemes can quickly restore traffic on permanent virtual circuits. These schemes are similar to self-healing FFP and Sonet rings but have more flexibility and better usage of spare capacity. Several standby virtual paths could be pre-established with zero capacity on different facilities to protect an existing virtual path on the primary route (Figure 3). Pre-assigning capacity for the standby virtual paths would be optional. The rerouting of virtual paths takes place at a logical layer via fixed backup routes over virtual path cross-connect systems (VPXs).

Protection control and path reconfiguration messages are carried by ATM's operations, administration and maintenance cells that could be transmitted at any time and could use bandwidth as needed. An ATM VPX can cross-connect virtual paths with any rate from zero to the line rate, more efficiently using transmission facilities.

ATM-based capacity management and dynamic reconfiguration have the potential to significantly reduce the transmission facilities required for network survivability, providing some economic justification for ATM's early deployment. Virtual path-level restoration using zero-bandwidth virtual paths-also known as hidden paths-on alternate routes could free up the restoration capacity, or a significant portion of it, for actual traffic growth, deferring significant capital expenditures that would otherwise be needed earlier.

Furthermore, the transmission cost per unit of bandwidth of the converted and future self-healing systems would be cut by up to 50%.

In this way, ATM and network survivability could help each other to become more affordable.

Hossein Ghandeharian is Section Manager of Network Planning and Implementation for Stentor Canadian Network Management, Ottawa, Ontario, Canada. His e-mail address is ghandeharian@stentor.ca.