In multi-hop wireless networks, longer-hop UDP and especially TCP flows receive a smaller fraction of the available bandwidth than single-hop flows in the absence of specific resource allocation techniques. Also, inter-node interference effects in multi-hop networks lead to incredibly poor bandwidth utilization. This paper begins to address these and other related issues through the novel concept of an investment function which captures the investments (network resources consumed, price paid for service quality, etc.) that are made in packet flows. The investment value of packets or packet flows can then be used to control their treatment in the network, with the goal of providing fair and efficient utilization of network resources through minimizing wasted investment and/or maximizing investment throughput. We show that a simple investment function can significantly improve flow fairness by smoothing out the disparities in distribution of bandwidth among flows with different hop length while simultaneously improving bandwidth utilization efficiency

For Internet telephony to be a viable alternative to the public switch telephone network (PSTN), efficient and high quality communications are required. This paper proposes an optimization algorithm that selects parameters like the coding scheme, packet loss bound, and maximum link utilization level in a voice over IP (VoIP) network The goal is to deliver guaranteed quality of service (for voice) while maximizing the number of users served. A VoIP architecture is also discussed that could use optimization algorithms to dynamically provision VoIP networks.

This paper advocates using the recently introduced index of variability (IDV) and a new measure, peak rate variability (PRV), to characterize the variability (burstiness) in real communications network traffic over the entire range of time scales. Further, we suggest the general hyperexponential interarrival distribution as a model suitable for network traffic and evaluate the ability of the third-order hyperexponential model to capture IDV, PRV and queuing characteristics. Although the hyperexponential interarrival distribution holds promise for network traffic modeling, in part due to its analytical tractability, we conclude that hyperexponential models with order larger than 3 will be required to model the burstiness of real network traffic adequately.

We provide a link between the concept of index of variability (IDV) and the multifractal wavelet model (MWM), that has been shown to capture the multifractal characteristics of network traffic. Index of variability has been proposed as an accurate measure of the degree of variability of network traffic (burstiness) over all relevant time scales. We introduce a new method that can be used to generate synthetic traffic traces from any valid IDV curve and the mean and variance of the associated network traffic trace.

This paper addresses the design and analysis of an end-to-end rate-based feedback flow control algorithm motivated by the available bit rate (ABR) service in wide-area asynchronous transfer mode (ATM) networks. Recognizing that the explicit feedback rate at time t will not affect the ABR buffer until time t+D for some D/spl ges/0, our approach is to first predict the ABR buffer status at time t+D, then base fuzzy-logic rate control decisions on these predicted values, and finally tune the controller parameters using gradient descent methods. Simulations show that this predictive self-tuning fuzzy-logic (PSTF) control scheme is efficient, stable, and outperforms other proposed ABR rate controllers in a variety of network environments. With delays corresponding to a US coast-to-coast connection, the PSTF controller can maintain high link utilization, avoid buffer overflows, and provide fair allocation of resources.

We evaluate the efficiency of AAL2 for voice transport in ATM networks using a simulation approach. After highlighting some of the general characteristics of AAL2, we adopt as the QoS requirement a bound on the K/sup th/ percentile delay in the AAL transmitter. We then determine the maximum number of homogeneous voice sources of a given bit rate that can be supported on a given bandwidth without violating this QoS requirement for AAL1, AAL2 and AAL5. Comparisons with simple analytical approximations are also provided. We conclude that AAL2 can support approximately 5 times as many 8 kb/s voice users as either AAL1 or AAL5, but all AALs can support approximately the same number of 64 kb/s users.

The rapidly deployable radio network (RDRN) is an architecture and experimental system to develop and evaluate hardware and software components suitable for implementing mobile, rapidly deployable, and adaptive wireless communications systems. The driving application for the RDRN is the need to quickly establish a communications infrastructure following a natural disaster, during a law enforcement activity, or rapid deployment of military force. The RDRN project incorporates digitally controlled antenna beams, programmable radios, adaptive protocols at the link layer, and mobile node management. This paper describes the architecture for the RDRN and a prototype system built to evaluate key system components.

In ATM networks, connection admission control (CAC) has been recognized as one of the most important means to provide satisfactory quality of service (QoS) and protect the network from congestion, especially for supporting real-time services such as voice and video. An efficient CAC strategy, which can both guarantee the QoS of admitted connections and achieve good resource utilization, then becomes a crucial issue for ATM network providers. The concept of a user-network traffic contract has been introduced by the ATM Forum. Starting from this point, we propose a measurement-based CAC strategy. We first discuss how to obtain an accurate description (traffic parameters) of user traffic by using on-line measurement in conjunction with dynamic renegotiation. The results show that the proposed strategy is reliable and simple to implement. We then move on to examine the possibility and methodology for exploiting the effect of statistical multiplexing in resource allocation to achieve higher network resource utilization.

Most algorithms proposed for controlling traffic prior to entering ATM networks are based on static mechanisms. Such static control mechanisms do not account for the dynamics of the user traffic or the network state. Some dynamic control algorithms have been proposed, but most of these algorithms are extremely complex and may make it difficult to provide real time control. In this paper, we present an adaptive rate control algorithm that has been implemented in hardware. The algorithm controls the traffic submitted by a source based on the indirectly observed average rate and burst size for the source. The algorithm is highly efficient and thereby provides real time control at high speed. Our implementation, in concert with flow control in the local area network, provides the basis for ATM-based high performance distributed systems.

This paper evaluates the concept of virtual path (VP) policing as an alternative to virtual channel (VC) policing in an asynchronous transfer mode (ATM) Ethernet network. The usage parameter control (UPC) function includes a peak rate and a sustainable rate policer. Both policers are implemented based on the leaky bucket algorithm defined by the ATM Forum. We implement VC and VP policers in parallel in a configuration containing variable bit rate (VBR) traffic VCs multiplexed in a VP. Simulations are used to measure the level of noncompliant traffic declared by the UPC function of both the VC and VP policing units. We evaluate the fairness of VP policing by having one group of VCs generate traffic in excess of their allocated bandwidth. We conclude that VP policing does not protect compliant VCs from non-compliant VCs. In fact, simulation results show that VP policing encourages cheating by penalizing compliant users more than non-compliant users.

This paper describes a gigabit LAN/WAN gateway being developed for the MAGIC gigabit testbed. The gateway interfaces a gigabit LAN developed by Digital's Systems Research Center and the MAGIC SONET/ATM wide area network. The gateway provides 622 Mb/s throughput between the LAN and WAN environments, and supports either a single STS-12c or four STS-3c tributaries. Traffic measurement capability and support for multiple bandwidth management schemes are provided by this architecture.>

The end-to-end cell discarding behavior of asynchronous transfer mode (ATM) telecommunication networks is considered. Cell discarding is performed in priority fashion by a nested threshold mechanism. Extensions to a previous single-queue analysis that allow the calculations of end-to-end discarding probabilities for each priority are presented. A computer tool based on this analysis, called DANTE (for discarding analysis in a network telecommunications environment) is described. It facilitates the calculation of end-to-end discarding probabilities for very general network topologies. This tool is used to generate sample results, illustrating important issues in designing networks to meet cell loss objectives. The computational efficiency of the computer-aided analytical tool is compared with that of simulation techniques. The presence of bursty traffic (not modeled here) is likely to change the cell discarding results quantitatively, but not the design issues identified herein.>

The authors describe the modeling and simulation of a frequency-division multiple-access (FDMA) satellite bandwidth-on-demand (BOD) service. The class of resource allocation processes to which the satellite BOD application belongs is identified and contrasted with more common resource allocation processes. Several unique features of BOD service complicate the performance specification, modeling, and analysis. The authors divide the modeling problem into three fundamental components: modeling the resource algorithm and modeling the processing of individual resource requests. These modeling components are illustrated with the satellite BOD application. In addition, simulation of the resource allocation requires three additional components: an easy-to-use user input interface, performance measures and statistics gathering, and presentation of simulation results (user output interface). The architecture of the simulation software (resource allocation analysis program, or RAAP) is presented in terms of these six components. To illustrate RAAP's capabilities, example simulation results are given.>

Overload control in asynchronous transfer mode (ATM) networks is dealt with via priority cell discarding mechanisms governed by a set of nested queue fill thresholds. An efficient search technique is developed which is applicable to any number of discarding priorities for determining the set of thresholds that optimize the offered load when constraints on cell loss (discarding) probabilities are specified. Results are presented as a function of queue capacity for given traffic distribution and arrival statistics. The results indicate that nested threshold discarding systems can support significantly higher traffic loads than systems without priorities. The sensitivity of the performance advantage to traffic mix and cell loss constraints is also illustrated. The efficient search procedure is shown to provide substantial computational gain when there are more than two discarding priorities.>

A problem associated with overload control by packet discarding in integrated packet networks such as a broadband integrated services digital network (B-ISDN) is considered. The question of optimality in priority packet discarding is examined in order to minimize the service degradation associated with the discarding. An analysis model which is consistent with asynchronous transfer mode (ATM) operation of such a network is presented. It is shown how dynamic programming can be used with this model to find an optimal discarding policy, and initial results of such an analysis for queuing systems with a single buffer position are presented. It is shown by example that the optimal policy can perform significantly better over a wide range of parameter values than the default policy of discarding arriving packets only when the queue is full.>

A replacement technique for lost speech packets is presented. This technique is based on the classification of the packets into four distinct classes: background noise, voiced speech, fricatives, and other. Different encoding schemes and lost packet replacement techniques are used for each class. Results of subjective tests indicate that giving preferential delivery treatment to packets based on class can be used to improve subjective quality. The replacement strategy renders the reconstructed signal indistinguishable from the original utterance for packet loss rates up to 47% for background noise packets, 8% for fricative packets, and 4% for other packets. For voiced packets, the replacement is distinguishable from the original signal even for packet loss rates lower than 5%, but significant improvement may be possible by reducing the memory associated with the voiced speech coding process.>

The authors discuss the control of short-term congestion, which is referred to as overload, in integrated packet networks (IPNs) containing a mix of data, speech, and possibly other types of signals. A system model is proposed that assigns a delivery priority to each packet (speech or otherwise) at the transmitter and discards speech packets according to delivery priority at any point in the network in response to overload. This model attempts to minimize per-packet processing at networks nodes. The research described is guided by two principles for IPN design: minimal per-packet processing and flexibility due to signal structure. The quality of the received speech is maintained by classifying speech segments according to their structure and coding them in a way that ensures ease of lost-packet regeneration at the receiver. The results of an experiment are reported that confirmed the general validity of this model from the standpoint of transmitter and receiver processing and subjective quality.>