San-qi Li

Congestion control for packet voice by selective packet discarding

In order to reduce the time delays as well as multiplexer memory requirements in packet voice systems, a family of congestion control schemes is proposed. They are all based on the selective discarding of packets whose loss will produce the least degradation in quality of the reconstructed voice signal. A mathematical model of the system is analyzed and queue length distributions are derived. These are used to compute performance measures, including mean waiting time and fractional packet loss. Performance curves for some typical systems are presented, and it is shown that the control procedures can achieve significant improvement over uncontrolled systems, reducing the mean waiting time and total packet loss (at transmitting and receiving ends). Congestion control with a resume level is also analyzed, showing that without increasing the fractional packet loss, the mean and variance of the queue can be reduced by selecting an appropriate resume level. The performance improvements are confirmed by the results of some informal subjective testing. >

Study of information loss in packet voice systems

Once a voice buffer is full, it remains full for a certain period, during which many packets are possibly blocked, resulting in consecutive clippings in voice. The packet loss rate during this period changes slowly and has large fluctuations. It is shown that the temporal behavior of packet loss, especially at high rate, is inherently determined by voice correlation and system capacity and is independent of buffer size. Buffering may reduce the occurrence of short blocking periods associated with low rates packet loss but does not affect long ones associated with high packet loss rates. In fact, increasing the buffer size merely extends nonblocking periods, and thereby reduces the overall average packet loss rate, but packet-loss performance within existing blocking periods is not significantly improved. A simple tool is developed for calculating the boundary performance. It is found that it is possible to design a packet-switched voice system without buffering only at the expense of supporting a fewer number of calls. The issue of voice delay allocation between source and network is discussed, and it is shown that it is more effective to keep the network delay short while extending the source delay. >

Overload control in a finite message storage buffer

An approach to the analysis of overload control in a finite buffer is introduced in which the original queuing process is modeled by a birth-and-death (BD) or quasi-birth-and-death (QBD) process. Overload control means to adapt the input process or the service process during the time period when the buffer content exceeds a certain level until it drops to another level. Such a control is necessary to reduce the occurrence of system shutdown periods and to protect high-priority messages against low-priority ones. Since the controlled process two be computed in terms of the will no longer be BD or QBD, the methodology commonly used for analyzing BD or QBD process cannot be applied. This makes direct analysis and computation of the controlled performance more complicated. The analytical methods consists in dividing the controlled process into two altering transient BD or QBD subprocesses, by observing only some selected transitions. Such a division enables the equilibrium probabilities of the controlled process to be computed in terms of the sojourn times of the two transient processes. It is shown that this is equivalent to the analysis and computation of equilibrium probabilities of the underlying stationary BD or QBD process. >

Performance of Voice/Data Integration on a TDM System

Data queueing is of primary concern in a voice/data integrated TDM system. The data queueing model is represented in the discrete-time domain with multiple servers and voice is given a higher priority than data. The data arrival process is assumed to be Poisson and the voice arrival process is characterized by a Markov chain. The correlation coefficient of the number of on voice calls between consecutive frames is used to measure the correlation behavior of the voice process. While the generating function approach may be used to analyze the queueing process, it involves the evaluation of a large number of boundary terms. On the assumption that the voice traffic consists of N i.i.d. two-state Markov chains, we derive a simple expression for the mean queue size as a function of two variables in the form of the traffic departure processes. The results clearly reveal a significant influence of the correlation coefficient on the data queueing process. Then, an approximate analysis based on the departure processes is introduced. The numerical and simulation results indicate that this approximate approach yields reasonably accurate results.

Performance trade-offs in an integrated voice/data services TDM system

Abstract A fairly tight upper bound for the mean queue size of a voice/data integrated TDM is derived. The result is applied to a queueing analysis of the data traffic subject to a prescribed voice call blocking probability. The analysis also indicates that the integration of voice and data services using time assignment speech interpolation (TASI) significantly improves the system performance over one without TASI. The problem of how to select the speech activity detection threshold at the voice source is addressed. We show that the voice correlation effect is heavily dependent upon the speech activity detection threshold at the voice source. A low threshold gives a low voice queueing delay and, therefore, a smaller time distortion on the received voice process.

A New Performance Measurement for Voice Transmission in Burst and Packet Switching

Voice transmission in burst switching is characterized by the process of talkspurt clipping, while in packet switching, it is characterized by the process of packet delay. In most analyses, the talkspurt clipping has been measured by the clipping probability averaged over all bits, and the packet delay has been measured by the delay performance averaged over all packets. The resulting measures overlook the duration of clipping in a talkspurt and the significant difference of delay in packets arriving at different times. Because of the nature of voice, different effects of these may result in substantially different degrees of voice distortion. This paper studies the worst case performance of both processes. The voice traffic is modeled as a process alternating between overload and underload periods. Statistically, more clipping and delay will be incurred while in the overload period. By worst case we mean that, in burst switching, we measure the worst case of talkspurt clipping duration in an overload period, while in packet switching, we measure the worst case of packet delay in an overload period. Furthermore, a simple closed form equation is derived which gives a very good approximation of the worst case mean packet delay performance. This equation can be more generally applied when the packet service time is to be geometrically distributed or when voice and data are to be integrated. The voice performances in burst switching and packet switching are also compared.

A study of traffic imbalances in a fast packet switch

The performance of a nonblocking space-division packet switch is studied given that traffics are imbalanced at input and output. Analysis shows that the performance of packet queuing delay at switch input, as well as the entire throughput of the switch, can be adversely affected by such imbalances. The work is then extended to examine a transient imbalance case, where the switch experiences the alternation of two transient periods, each at a different traffic imbalance mode. The alternation is modeled by a two-state Markov chain. Both balanced and imbalanced cases can be viewed as the two extremes of the transient case. It is observed that the system throughput and the queuing performance in the transient case heavily depend on both mean sojourn time and steady-state probability at each imbalance mode.<<ETX>>

Performance of a nonblocking space-division packet switch in a time variant nonuniform traffic environment

The authors study the performance of a nonblocking space-division packet switch, given that the traffic intensities at the switch not only are nonuniform but also change as a function of time. A finite-state Markov chain is used as an underlying process to govern the time variation of traffic for the entire switch. The packet arrivals at each input form an independent Bernoulli process modulated by the underlying Markov chain. The output address of each packet is independently and randomly assigned with probability distributions, which are also modulated by the Markov chain. Provided that the traffic on each output is not dominated by individual inputs the service time of each output queue for sufficiently large switches can be characterized by an independent Markov modulated phase-type process. A matrix geometric solution for the resultant quasi-birth-death type queuing process is presented. The maximum throughput is obtained at the system saturation. The performance of the switch is numerically examined under various traffic conditions. A contention priority scheme to improve the switch performance is proposed. >

Simulation study of a network of voice/data integrated TDMs

Previous results on a single integrated-services TDM (time-division multiplexer) node are extended to the modelling and analysis of a network of integrated-service TDMs. Different interconnections of the integrated-services TDM nodes are studied by means of computer simulation. The results indicate that a Poisson assumption for the internal data arrivals is reasonable. With this assumption, the mean queue analysis at an individual node can be separated from the rest of the network, so that solution of the entire network can be obtained by combining the separate solutions. >

Traffic characterization for integrated services

A traffic model in which the interruption traffic is a correlated process is introduced. In this model the effect of correlation is culminated in a parameter that is the sum of all the imbalances of the autocorrelation function of the interruption process. Based on this model, the queueing analysis of integrated services on a packet switched TDM system is developed. It is shown that the presence of correlation increases the mean queue length. An adaptive flow control scheme that successively adjusts the short-bursty traffic arrival rate at each TDM node to reduce the correlation effect on the queue length build-up is proposed.<<ETX>>