ATM Traffic Control

The control of ATM traffic is complicated as a result of ATM’s high-link speed and small cell size, the diverse service requirements of ATM applications, and the diverse characteristics of ATM traffic. Furthermore, the configuration and size of the ATM environment, either local or wide area, has a significant impact on the choice of traffic control mechanisms.

The factor that most complicates traffic control in ATM is its high-link speed. Typical ATM link speeds are 155.52 Mbit/s and 622.08 Mbit/s. At these high-link speeds, 53- byte ATM cells must be switched at rates greater than one cell per 2.726 ms or 0.682 ms, respectively. It is apparent that the cell processing required by traffic control must perform at speeds comparable to these cell-switching rates. Thus, traffic control should be simple and efficient, without excessive software processing.

Such high speeds render many traditional traffic control mechanisms inadequate for use in ATM because of their reactive nature. Traditional reactive traffic control mechanisms attempt to control network congestion by responding to it after it occurs and usually involves sending feedback to the source in the form of a choke packet.

However, a large bandwidth-delay product (i.e., the amount of traffic that can be sent in a single propagation delay time) renders many reactive control schemes ineffective in high-speed networks. When a node receives feedback, it may have already transmitted a large amount of data. Consider a cross-continental 622 Mbit/s connection with a propagation delay of 20 ms (propagation-bandwidth product of 12.4 Mbit).

If a node at one end of the connection experiences congestion and attempts to throttle the source at the other end by sending it a feedback packet, the source will already have transmitted over 12 Mb of information before feedback arrives. This example illustrates the ineffectiveness of traditional reactive traffic control mechanisms in high-speed networks and argues for novel mechanisms that take into account high propagation-bandwidth products.

Not only is traffic control complicated by high speeds, but it also is made more difficult by the diverse QoS requirements of ATM applications. For example, many applications have strict delay requirements and must be delivered within a specified amount of time. Other applications have strict loss requirements and must be delivered reliably without an inordinate amount of loss. Traffic controls must address the diverse requirements of such applications.

Another factor complicating traffic control in ATM networks is the diversity of ATM traffic characteristics. In ATM networks, continuous bit rate traffic is accompanied by bursty traffic. Bursty traffic generates cells at a peak rate for a very short period of time and then immediately becomes less active, generating fewer cells. To improve the efficiency of ATM network utilization, bursty calls should be allocated an amount of bandwidth that is less than their peak rate.

This allows the network to multiplex more calls by taking advantage of the small probability that a large number of bursty calls will be simultaneously active. This type of multiplexing is referred to as statistical multiplexing. The problem then becomes one of determining how best to multiplex bursty calls statistically such that the number of cells dropped as a result of excessive burstiness is balanced with the number of bursty traffic streams allowed. Addressing the unique demands of bursty traffic is an important function of ATM traffic control.

For these reasons, many traffic control mechanisms developed for existing networks may not be applicable to ATM networks, and therefore novel forms of traffic control are required (8,9). One such class of novel mechanisms that work well in high-speed networks falls under the heading of preventive control mechanisms. Preventive control attempts to manage congestion by preventing it before it occurs. Preventive traffic control is targeted primarily at real-time traffic. Another class of traffic control mechanisms has been targeted toward non-real-time data traffic and relies on novel reactive feedback mechanisms.