Broadband networks based on asynchronous transfer mode (ATM) have to support traffic with widely different traffic characteristics and quality of service requirement. In this paper, we extend our earlier work in and develop a dynamic connection admission controller (CAC) that supports cell loss requirements. The CAC algorithm explicitly computes the equivalent bandwidth required to support each class of connections based on on-line observations of aggregate traffic statistics as well as the declared parameters. We use Gaussian and diffusion approximations to characterize the aggregate traffic stream, and use fuzzy control strategy to combine model and measurement results to derive simple closed-form formulas to estimate the equivalent bandwidth in real time. We validate the proposed algorithms for various variable bit-rate traffic profiles and show that the system utilization can be substantially improved by appropriately tuning the fuzzy logic controller to combine traffic characteristics deduced from the declared parameters and traffic measurements.

We present a control-theoretic approach to the design of closed-loop rate-based flow control in high-speed networks. The proposed control uses a dual proportional-plus-derivative controller, where the control parameters can be designed to ensure the stability of the traffic patterns and propagation delays. We show how the control mechanism can be used to design a controller to support ABR service based on feedback of explicit rates (ERs). We demonstrate the excellent transient and steady-state performance of the controller through a number of examples. We also show experimental results that have been obtained from our asynchronous transfer mode (ATM) testbed, which consists of two interconnected ATM LANs, one located in Princeton, NJ, and the other in Berlin, Germany, with an all-software ER-controller implementation.

This paper presents a robust and flexible real-time scheme for determining appropriate parameter values for usage parameter control (UPC) of an arbitrary source in an asynchronous transfer mode network. In our approach, the UPC parameters are chosen as a function of the statistical characteristics of the observed cell stream, the user's tolerance for traffic shaping, and a measure of the network cost. For this purpose, we develop an approximate statistical characterization for an arbitrary cell stream. The statistical characterization is mapped to a UPC descriptor that can be negotiated, with the network. The selected UPC descriptor is optimal in the sense of minimizing a network cost function, subject to meeting user-specified constraints on shaping delay. The UPC estimation scheme is extended to adapt to slow time-scale changes in traffic characteristics via dynamic renegotiation of the UPC parameters. We illustrate the effectiveness of our methodologies with examples taken from MPEG video sequences.

This paper proposes a growable switch architecture with multiple modules comprising a CORE module to support highspeed switching and EDGE modules with large buffers, providing a large number of traffic classes. For open-loop traffic such as CBR and VBR, the proposed switch uses a weighted fair queueing scheduler at each EDGE buffer module based on the equivalent bandwidth allocated by the call admission control (CAC). On the other hand, for best effort class services (e.g., ABR and UBR), closed-loop control using adaptive rate control at the actual source and at the virtual source queue of the input EDGE module is proposed to avoid internal congestion and to achieve fair throughput performance among the flows competing on the same link. Simulation results show that the proposed architecture achieves ATM switches having hundreds of Gbps capacity, while providing a large number of QoS classes with high throughput and fair sharing of bandwidth by all flows.

We present a control theoretic approach to the design of closed loop rate based flow control in high speed networks. The proposed control uses a dual proportional derivative controller, where the control parameters can be designed to ensure the stability of the control loop in a control theoretic sense, over a wide range of traffic patterns and propagation delays. We show how the control mechanism can be used to design a controller to support available bit rate (ABR) service based on feedback of explicit rates. We demonstrate the excellent transient and steady state performance of the controller through a number of examples.

In high-speed networks based on asynchronous transfer mode (ATM), variable bit rate (VBR) sources generate bursty cell streams which share the link bandwidth wherever multiplexing occurs. In theory, the bandwidth requirement per stream to support a given quality-of-service at a multiplexer should generally decrease as the number of streams increases. In practice, high statistical multiplexing gain is difficult to achieve because good traffic characterizations are usually not available. This paper proposes the use of peakedness-based measures to capture the statistical information from traffic streams needed to determine bandwidth allocations. The standard definition of peakedness applies only to point process models of traffic. Yet fluid models of traffic have certain advantages in terms of tractability. Hence, we introduce a new measure called modified peakedness, which encompasses point process and fluid models under a common framework. We develop some of its properties and specialize it to several common traffic models. Finally, we study the effectiveness of the peakedness/modified peakedness as burstiness measures for real-time traffic.

Broadband networks based on ATM have to support multiple classes of services with widely different traffic characteristics and quality of services. We propose and develop a multi-class connection admission control (CAC) policy that supports cell loss and delay requirements. In this model-based CAC, the source traffic is described in terms of the usage parameter control (UPC) parameters. Through careful analysis and approximations, we derive simple closed-form formulas to calculate the bandwidth required to meet guarantees on quality of service (QoS). While being robust, the CAC achieves a high level of resource utilization and can be easily implemented for real-time admission control.

We describe the fuzzy explicit fate marking (FERM) traffic flow control algorithm for a class of best effort service, known as available bit rate (ABR), proposed by the ATM Forum. FERM is an explicit rate marking scheme in which an explicit rate is calculated at the asynchronous transfer mode (ATM) switch and sent back to the ABR traffic sources encapsulated within resource management (RM) cells. The flow rate is calculated by the fuzzy congestion control (FCC) module by monitoring the average ABR queue length and its rate of change, then by using a set of linguistic rules. We use simulation to compare the steady-state and transient performance of FERM with EPRCA (a current favourite by the ATM Forum) in the presence of high priority variable bit rate (VBR) video and constant bit rate (CBR) in both a local-area network (LAN) and a wide-area network (WAN) environment. Our experiments show that FERM exhibits a robust behavior, even under extreme network loading conditions, and ensures fair share of the bandwidth for all virtual channels (VCs) regardless of the number of hops they traverse. Additionally, FERM controls congestion substantially better than EPRCA, offers faster transient response, leads to lower end-to-end delay and better network utilization.

In networks based on the asynchronous transfer mode (ATM), traffic sources are subjected to a usage parameter control (UPC) function which controls the emitted cell stream by either discarding or delaying cells according to the parameters of a deterministic algorithm (e.g. the dual leaky bucket). The UPC parameters constitute a traffic descriptor which are used to allocate network resources to the traffic source. This paper presents a scheme for determining dual leaky bucket UPC parameter values for an arbitrary traffic source from observations of its emitted cell stream. The selection of UPC parameters is based on a novel statistical characterization of the observed cell stream, the user's tolerance for traffic shaping and a cost function based on the network resource allocation policy. Our scheme makes weak modeling assumptions and is robust for a wide class of traffic streams.

The paper presents a model of a network with multiple controls that are hierarchically arranged in time, based on the time constants of events they control. The authors model the network with such hierarchical controls as a closed multi-chain queueing network and use a quasi static approximation based on a hierarchical decomposition technique to solve the model. They determine the blocking probability at the call level and burst level, as well as the waiting time distribution for bursts that are blocked. They investigate the trade off in performance that can be achieved at different levels in the hierarchical control model.>

The notion of perceptible degradation of quality of the received signal seen by an individual user is introduced. It is proposed that the design of high-speed networks that support real-time traffic be carried out on the basis of a new performance measure, which is defined as the expected number of perceptible degradations that a customer is likely to encounter per unit of time. By using fluid-flow models, an analytical methodology is obtained for calculating the new performance measure. The method is illustrated by representative numerical examples from voice and video traffic for various link speeds and buffer sizes.>

The authors present a predictive congestion control policy for carrying data traffic in the presence of real-time traffic that has stringent delay requirements, such as voice and video traffic. The predictive policy attempts to predict the buffer content of a node at some time in the future. The policy is based on information exchange between adjacent nodes of the network. The cross traffic is characterized by its mean rate, its variance, and the correlation, and the performance of the scheme is studied over a wide range of cross traffic parameter values. The effect of higher link speeds and larger propagation delays on the performance is studied. The predictive policy is compared to other policies. The investigation shows that predictive control policy performs significantly better than static or adaptive end-to-end rate control policies.>

The authors develop a new analytic model for a variable bit rate video source, which can model a variety of video applications by a suitable choice of model parameters. Using the model they studied the performance of a video multiplexer that multiplexes several full motion video sources. The video sources were assumed to use a layered coding technique. A service strategy is proposed that guarantees delivery of packets marked essential, whereas nonessential packets may be dropped in the event of congestion. The study showed that with layered coding, good tail probabilities can be obtained for a given number of sources at the expense of some loss for nonessential packets. The performance of the video multiplexer was investigated when the video sources were correlated, and when independent video sources with different thresholds for essential packet rate were multiplexed together.>

The authors examine the networking environment of the future, pose the technical questions that need to be addressed to provide service assurance, and propose a set of strategies for congestion avoidance. Congestion avoidance is recommended instead of congestion control, which has very limited value in a high-speed network, where the latency in detecting congestion and reacting to it may make the control ineffective. The proposed solution consists of enforcing rate-control at the network edges, bandwidth reservation for continuous bit oriented (CBO) traffic using a unique two-queue strategy, and a scheduler-based packet cross-connect system. Simulation results of a Q+ model of an access node are presented that quantify the performance of this solution and compare it to existing congestion control mechanisms. The results demonstrate that it is possible to meet the differing needs of traffic types needing datagram transport and those needing virtual-circuit transport by using an intelligent scheduling strategy and a meaningful packet transport protocol.>

A packet multiplexer is modeled for continuous bit rate (CBR) traffic in an asynchronous transfer mode (ATM) network as an nD/D/1 queue. The efficiencies of various algorithms for finding the delay distribution are compared. In particular, a new algorithm is proposed whose time complexity is O(n/sup 2/), where n is the number of voice sources being multiplexed. The use of the central limit theorem can reduce the time complexity to O(n) for large n. An asymptotic formula is found whose time complexity is independent of n and it works well (for practical purposes) over a wide range of parameter values. The authors examine and comment on the use of the M/D/1 results as an approximation. In addition to comparing the performances of these algorithms, they show that the buffer requirements for such a queue are significantly less than the theoretical maximum (even when the requirement on the call disruption probability is very low). This result has important implications in the design of buffer size. The buffer requirement is relatively insensitive to the design criterion (call disruption probability).>

The authors propose a congestion control strategy called distributed source control (DSC), designed to address the integration of diverse traffic types in broadband packet networks. DSC is a rate-based network access control implemented at the asynchronous transfer mode (ATM) layer as a feedforward control, in conjunction with an adaptive end-to-end control between network edges. The authors examine the performance of high-speed data traffic such as large file transfers, still images, and document retrievals. The authors study DSC's integrability with real-time traffic such as voice and video. With the help of a simulation model of a broadband packet network (150 Mb/s), the authors quantify the improvement in network performance due to DSC. An analytic model for an access node under DSC is developed, and guidelines for sizing of buffers in the switch and for choosing the end-to-end window size are provided.>

A multidimensional approach to congestion control in integrated broadband networks is described, and a framework for a congestion control strategy that is applicable to virtual-circuit-based services as well as connectionless services is developed. The proposed congestion control strategy is based on a set of temporal controls, classification of services is based on their traffic characteristics and quality of service (QOS) requirements, and partitioning of the control space is based on the reaction time of the controls and the network time constants. The congestion control framework is simple, scales as the delay-bandwidth product increases, and is well suited for the asynchronous transfer mode that is being promoted as the underlying transport for B-ISDN.>

The performance of a statistical multiplexer whose inputs consist of a superposition of voice packet streams is studied. The delay for such a system is analyzed by solving the Sigma D/sub i//D/1 queue. The analytic method can be used to find the approximate mean delay for an arbitrarily large number of trunks and the approximate delay distribution when the number of trunks is less than 100. An efficient hybrid simulation of the packet voice multiplexer which can be used to find the delay distribution for a large number of trunks is presented. In addition, easily computable error bounds for the present approximation are provided, and the accuracy of the M/D/1 approximation is investigated.>

The authors propose a congestion control strategy called distributed source control (DSC), which is designed to address the integration of these diverse traffic types in broadband packet networks. An examination is made of the performance of high-speed continuous-stream data traffic such as large-file transfer and still-image and document retrievals, which are projected to be high runners in future networks and for which congestion control in broadband systems is a serious problem. Their integrability with real-time traffic such as voice and video is studied. With the help of a simulation model of a 150-Mb/s broadband packet network, the improvements in the performance of the different traffic types by using DSC are quantified. The authors also provide guidelines for sizing buffers in the switch and recommend a call-acceptance policy for these traffic types in the network.>

With the advent of high-speed networks, there has been a tendency to simplify flow/congestion control functions at lower levels of the OSI model and relegate them to higher levels. This paper shows that a class of traffic, namely host-to-host file transfers, can experience severe degradation in performance due to lack of proper control in such networks. A simple distributed rate-based flow control scheme is proposed which solves the problem without affecting the performance of the network. The scheme is easily implemented, with control totally distributed and managed by the sources. The rate is negotiated at call setup time and entails very little overhead. A detailed simulation model is built to quantify the problems and verify the solution.>