ATM Switch Traffic Management Essentials Primer from Integrated Telecom Technology. Source Behavior for ATM ABR Traffic Management: An Explanation IEEE Communications, 11/1/96. The Performance of TCP Over ATM ABR and UBR Services By the authors of the above. Comparison of Fast Resource Management and Available Bit Rate Techniques for Efficient use of ATM Bandwidth Describes and compares the implementation of the Fast Resource Management and ABR techniques in a European trial. In Postscript. ATM Forum's Traffic Management 4.0 specification In PDF - requires Adobe Acrobat to read. Policy-based QoS in Ethernet networks Policy-based QoS in Ethernet networks Network World, 6/9/97. Guy Trotter is a product marketing engineer for Hewlett-Packard Co.'s Communications Measurements Division in Edmonton, Alberta. He can be reached at guy_trotter@ hp.com

By Guy Trotter
Network World, 9/15/97

One of ATM's greatest advantages is its built-in ability to support user-specified quality of service (QoS) levels.

To deliver QoS, the ATM Forum has defined five service classes, including available bit rate (ABR) service.

ATM allows users (or applications) to obtain end-to-end connections at varying bandwidth levels, making ABR well suited for the flexible transfer of data, voice and video. Defined service classes allow users to trade off bandwidth efficiency, delay and potential cell loss to achieve an acceptable QoS level.

For example, unspecified bit rate (UBR) service promises best-effort delivery but may be subject to high cell loss rates and unpredictable network management capabilities. Constant bit rate (CBR) provides minimal delay but requires that a fixed amount of bandwidth be allocated for the entire session. This is an appropriate approach for voice but inefficient for many data applications.

Variable Bit Rate (VBR) services come in two flavors: real time and non-real time. They offer somewhat more flexibility with service guarantees, but users must specify in advance a range of parameters that describe their potential traffic ''profile,'' such as peak and sustained cell rates or maximum burst sizes.

Unfortunately, it is difficult to strictly characterize many of today's unpredictable applications within these parameters because there may be long periods of low data transfer followed by large bursts of data.

Moreover, if actual traffic falls outside the profile - as unpredictable bursty traffic often will - then cells may be discarded, resulting in the corruption of application layer information and a sharp decrease in throughput.

What is needed, then, is a bandwidth-efficient, low cell loss service with the flexibility to handle large data bursts.

Enter ABR.

ABR provides a mechanism for dynamic allocation of available bandwidth equitably among ATM network users. In many respects, ABR represents a best-of-both-worlds offering for applications such as Web services because it accommodates varying traffic profiles without being a bandwidth hog.

ABR operates as a feedback mechanism that utilizes special ATM cells, called Resource Management (RM) cells. RM cells carry requests through the network to determine if higher transmission rates can be supported and deliver an answer back to the originating user.

Here's how: The user begins by transmitting data cells at an initial rate, interspersed with RM cells. If the source wishes to increase its transmission data rate, it indicates the request via an RM cell transmitted into the network.

The RM cell is then processed and passed progressively through each switch in the connection. Every switch examines the RM cell and determines whether it can support the requested rate increase, taking into account a fair allocation of bandwidth among all other active ABR sources.

Each switch can choose one of three options. If the switch can support the request, the RM cell is passed along unchanged. If the switch is experiencing congestion but can still accommodate the current user traffic rate, it modifies the RM cell to indicate that the source should not change its rate of transmission. If the switch is experiencing congestion that is slowing the network, it can modify the RM cell to indicate that the source should reduce its user traffic rate. Or the switch can tell it to transmit at a specific data rate.

The RM cell is then passed on to the next switch for a similar evaluation until it reaches its ultimate destination. If any switch has indicated that the source must maintain, or decrease, its traffic rate, then no switch can modify the RM cell to indicate an increase.

Finally, the RM cell is looped back to the source, which takes the specified action. It increases, maintains or decreases its rate of transmission or resets its rate to the level specifically indicated by the RM cell.

As long as the user transmits at the approved rate, cell loss rates are guaranteed to be low. Additional bandwidth can be dynamically requested and obtained, maintaining the low cell loss rate.

On the other hand, ABR also helps manage overall network resources by instructing users to reduce transmission rates if needed and releasing bandwidth to the network for use by other sources.

The result is increased application and network efficiency. Because cells are transmitted only at rates that can be handled by the network, fewer cells are lost, requiring fewer frames to be re-transmitted, and bandwidth is allocated more equitably among users.

Most ATM equipment manufacturers currently are implementing ABR support in their switches. Based on normal testing, evaluation and deployment cycles, it is reasonable to expect that network users should begin to see ABR services by mid-1998.

Given the growth of applications with varying nonreal-time transmission requirements, ABR should become a key building block for ATM network managers in the future.