'p$.aEc Modeling in a Multi-media Environment

In this paper we propose a new model for characterizing the data traffic in a multi-media environment. We model the data traffic by a two- state doubly stochastic Poisson process, with SO- journ times in each state having an independent and identical heavy tided distribution. such as the hreto distribution. The simulation results from the new data traffic model are presented. The new model is versatile in capturing the self-similar characteristics of traffic found in the recent traf- fic measurements. We also suggest that the other two types of multi-media traffic namely, voice and video may each be characterized by a 2-state dou- bly stochastic Poisson process with exponential so- journ times (Le-, a Markov modulated Poisson pro- cess or MMPP). I. INTRODUCTION With the advent of B-ISDN, significant effort has been devoted to supporting real time traffic such as voice and video along with jitter tolerant traffic such as data traffic, in a packet switched environment. These wide spectrum of traffic sources (such as computer data, VBR video, voice. etc.,) exhibit a diverse mixture of traffic characteristics, Hence it is imperative to develop a model that aptly char- acterizes the variabilty and statistical correlations of the packet arrival process. This model may then be used to evaluate the network performance (QOS, utility,etc.) or to evaluate the connection admission control and source policing algorithms. Various models have been introduced in the literature, for the characterization of these sources. One such model is the ON-OFF source model, which has been successfully used to characterize traffic from a single voice source. The ON-OFF source model has three parameters Q. b and T and switches between 2 states: an active or (ON) state or (OFF) state when it generates no packets. The ON and OFF periods are exponentially distributed with parameters a and 6 respectively. Video traffic, which is anticipated to occupy most of the bandwidth in the future broadband networks, is highly bursty. The characteristics of VBR video depend upon many aspects like information content of the video (eg. picturephone, broadcast television, teleconference, etc.). Hence, modeling a VBR video is an even more complex task since, the bit rate possesses a high degree of variabil- ity at different levels, namely at a scene level, frame level w-hen it generates packets at a constant rate r and a srlenf or intra frame level. For video sources, with uniform aceiv- ity level, the continuous-time, discrete-state hlarkov model proposed by (l), consisting of the superposition of many mini ON-OFF sources is an appropriate model. Conventional models of data traffic are Poisson, Batch Poisson, Markov modulated Poisson process or Fluid F'low models. However recently measured results (2) (3) (4) (a) (SI, indicate the inadequacy of these standard traffic models and stress the requirement for a more refined model. These measured results indicate that data traffic is statistically self-similar (Le., it exhibits a high level of variability over a wide range of time scales). Hence new models that can express self-similar (or fractal) characteristics have been Also, earlier measurements of data traffic (IO) indicate that the message length distribution is bimodal. Since a burst of packets are produced for each message, this also suggests that the burst of data packets may be bimodally distributed. Thus the data traffic may consist of short and long bursts. Based on these observations, we propose a new model for data traffic. We model the data. traffic by a 2- state doubly stochaseic Poisson process: with sojourn times in each state having an independent and identical heavy tailed distribution, such as the Pareto distribution. The two states of the switched Poisson process may correspond to the long and short burst rates. This model captures the long range dependence present in data traffic. Though each type of traffic has been characterized by suitable models, there is a need for an aggregate model that could be used either as an easy means for obtain- ing the performance of the multiplexer to which the traf- fic is fed or for simulating the arrival process easily. Due to the co-existence of fractal and non-fractal traffic in the multi-media environment: the task of modeling the traffic mix poses a great challenge to the modeler. In the few aggregate models proposed (11) (12) (In) (11) (I51 (IS) for multi-media traffic, the self-similar characteristics of the component traffic has not been accounted for. htorivated by this fact we suggest a new aggregate model consisting of switched Poisson processes. To this end we suggest that the aggregate of voice sources and aggregate of video sources be approximated by a M51PP. Data traffic stream. which is self-similar may be modeled by the new model proposed. The resulting aggregate model is simple and easy to simu- late.