## The Effect of Multiple Time Scales and Subexponentiality on the Behavior of a Broadband Network Multiplexer (1996)

### Cached

### Download Links

Citations: | 57 - 17 self |

### BibTeX

@MISC{Jelenkovic96theeffect,

author = {Predrag R. Jelenkovic and Predrag R. Jelenkovi'c},

title = {The Effect of Multiple Time Scales and Subexponentiality on the Behavior of a Broadband Network Multiplexer},

year = {1996}

}

### OpenURL

### Abstract

The Effect of Multiple Time Scales and Subexponentiality on the Behavior of a Broadband Network Multiplexer Predrag R. Jelenkovi'c The main theme of this dissertation is the evaluation of the capacity of broadband multimedia network multiplexers. This problem calls for the modeling of network traffic streams and the analysis of a network multiplexer that is loaded with the corresponding models. For modeling we focus on MPEG video traffic streams that are expected to be predominant in the traffic mixture of future multimedia networks. We experimentally demonstrate that real-time MPEG video traffic exhibits multiple time scale characteristics, as well as subexponential first and second order statistics. Then we construct a model of MPEG video that captures both of these characteristics and accurately predicts queueing behavior for a broad range of buffer and capacity sizes. Depending on whether a network multiplexer (loaded with MPEG) is strictly or weakly stable the dominant effect o...

### Citations

453 | Branching processes
- Athreya, Ney
- 1972
(Show Context)
Citation Context ...ibution [34]: , and appropriately chosen. In what follows, we will state two basic results on subexponential distributions. The general relation between and is the following. Lemma 1: (Athrey and Ney =-=[35]-=-) . Lemma 2: If then as , for all . Note: Lemma 2 clearly shows that for long-tailed distributions, Cramér-type conditions are not satisfied. APPENDIX C WEAK STABILITY: PROOF OF THEOREM 2 In this sect... |

419 |
Equivalent capacity and its application to bandwidth allocation in high-speed networks
- Guerin, Ahmadi, et al.
- 1991
(Show Context)
Citation Context ...to guarantee a given quality of service while efficiently utilizing resources, one wants to accurately estimate the tail of the queue, and derive from that the traffic stream’s “equivalent bandwidth” =-=[27]-=-. However, a simple-minded application of the notion of equivalent bandwidth based only the decay rate of the queue tail would, in this case, overestimate the queue by a factor of 1 (i.e., two orders ... |

270 | Statistical properties of MPEG video traffic and their impact on traffic modeling
- Rose
- 1995
(Show Context)
Citation Context ...VIDEO STREAMS A. MPEG Video Data The video data employed consisted of video sequences of MPEG-I frame sizes, created at the Institute of Computer Science, University of Würzburg, and are described in =-=[19]-=-. In all, 17 sequences (sportscasts, movies, music videos, newscasts, talk shows, cartoons, and “set top”) of 40 000 frames each were available. Taking as input the raw video frames, the MPEG coder pr... |

268 | Effective bandwidth of general Markovian traffic sources and admission control of high speed networks
- Elwalid, Mitra
- 1993
(Show Context)
Citation Context ...amily is in ), and that the Cramér conditions are satisfied. Then, assuming that the queue is weakly stable Proof: Given in Appendix C. Remarks: 1) This result strongly supports the fluid models (see =-=[29]-=-) that have much lower computational complexity; 2) related asymptotic results in the Markovian framework that justify the fluid approximation were obtained in [9] and [30]; 3) this result can be prov... |

243 |
Introduction to Stochastic Processes
- Cinlar
- 1975
(Show Context)
Citation Context ...ime indicator process has Markovian jumps to the next state, but arbitrarily distributed sojourn times in each state. Thus, the regime indicator process is a Markov renewal process (see, for example, =-=[21]-=-). To finish the construction of the model, we define four mutually independent i.i.d. processes , with marginal distribution , mean , each being independent of and . Each process models the fast time... |

216 |
The stability of a queue with non-independent inter-arrival and service
- Loynes
- 1962
(Show Context)
Citation Context ...ess and constant server capacity , whose queue size at time is defined by Lindley’s recursion (5) For the rest of the paper, we assume that is the unique stationary solution to the recursion (5) (see =-=[26]-=-), i.e., we assume that the queue is in its steady state. The arrival process , represents the mathematical model for MPEG video streams as described in Section II-C. In this section, we consider the ... |

154 | Traffic Modeling for Telecommunications Networks - Frost, Melamed - 1994 |

116 | The importance of long-range dependence of VBR video traffic
- Ryu, Elwalid
- 1996
(Show Context)
Citation Context ...rom the literature. Indeed, a number of authors [3], [15], [16] contend that long-range dependence of one form or another has a dominant impact on the queue, while others claim that it does not [17], =-=[18]-=-, [4]. Our framework allows one to synthesize these results1054 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 15, NO. 6, AUGUST 1997 in that the former can be viewed as instances of the weak... |

114 | Queue Response to Input Correlation Functions: Discrete Spectral Analysis
- Li, Hwang
- 1992
(Show Context)
Citation Context ...el properties of video such as scenes (seconds), to entire movies or video calls. Different modeling perspectives have been taken in examining this complex time-dependency structure. Li and Hwang, in =-=[3]-=-, argue from the frequency domain point of view that the low frequency band of the autocorrelation’s Fourier transform (long-term correlation) has the most significant impact on queueing. Lazar et al.... |

105 | Fundamental bounds and approximations for ATM multiplexers with applications to video teleconferencing
- Elwalid, Heyman, et al.
- 1995
(Show Context)
Citation Context ... and realistic class of AFC’s can be matched, including, for example, Markovian models such as the Markov-modulated Poisson process of Skelly et al. [17], the TES-based models [22], or the DAR models =-=[24]-=-. Note that all of these examples taken from the literature can only match exponentially decreasing ACF’s. Third, an SRP model can match subexponential ACF’s, and as such, it is a uniquely efficient a... |

99 |
Source Models for VBR Broadcast-Video Traffic
- Heyman, Lakshman
- 1996
(Show Context)
Citation Context ...[11] investigated the long-range dependency of the autocorrelation function of video conference traces. We observed a subexponential scene length duration in MPEG video streams (this was also seen in =-=[12]-=- for video coded by a simpler DPCM scheme without motion compensation), which is responsible for the long-range dependency. A calculus of almost negligable computational complexity (unlike the case of... |

86 | Analysis of an ATM Buffer with Self-Similar (“fractal”) Input Traffic
- Likhanov, Tsybakov, et al.
- 1995
(Show Context)
Citation Context ... in the strict stability scenario, the dominant effect is due to the fast time scale buildups—put in the broader context some conflicting results from the literature. Indeed, a number of authors [3], =-=[15]-=-, [16] contend that long-range dependence of one form or another has a dominant impact on the queue, while others claim that it does not [17], [18], [4]. Our framework allows one to synthesize these r... |

70 | Asymptotic results for multiplexing subexponential on-off processes
- Jelenković, Lazar
- 1999
(Show Context)
Citation Context ...ever, there is no explicit algorithm to extract model parameters from the second-order statistics of an empirical trace. (Precise fluid queue asymptotics with arrivals were obtained in [13], [2], and =-=[25]-=-.) Fourth, in the presence of long-tailed renewal times, Theorem 2 (see Section IV-B) further justifies the usage of the fluid type models. We now use this SRP model to generate traffic matching that ... |

56 |
Subexponential distributions and integrated tails
- Klüppelberg
- 1988
(Show Context)
Citation Context ...st introduced by Chistakov [32]. Some examples of distribution functions in are the following. 1) Distributions of regular variations (contains Pareto distributions) 5) Benktander Type I distribution =-=[34]-=- and appropriately chosen. 6) Benktander Type II distribution [34]: , and appropriately chosen. In what follows, we will state two basic results on subexponential distributions. The general relation b... |

45 | Subexponential asymptotics of a Markovmodulated random walk with queueing applications
- Jelenković, Lazar
- 1998
(Show Context)
Citation Context ...ueing analysis. Accordingly, we design a video model that accurately reflects these characteristics. The model is structured so that it is amenable to recently developed queueing theoretical analysis =-=[1]-=-, [2]. In the literature, the existence of multiple time scale statistics in video traffic has been consistently observed by many authors. The bit rate of a video stream exhibits dependencies (correla... |

38 |
Modeling Video Sources for Real-Time Scheduling
- Lazar, ci, et al.
- 1993
(Show Context)
Citation Context ... argue from the frequency domain point of view that the low frequency band of the autocorrelation’s Fourier transform (long-term correlation) has the most significant impact on queueing. Lazar et al. =-=[4]-=- developed video models for the slice and frame time scales, and showed that in the case of strict quality of service (QOS) requirements (small time delay), precise modeling of the high-order autocorr... |

34 | Statistical multiplexing of multiple time-scale Markov streams
- Tse, Gallager, et al.
- 1995
(Show Context)
Citation Context ..., it was analytically shown that when a stream with multiple time scales passes through a queue, the queue length distribution has multiple decay rates. Similar results were independently reported in =-=[9]-=-. Furthermore, a perturbation theory technique that exploits the multiple time scale structure of the arrival process and leads to a computationally efficient algorithm for evaluating the queue length... |

31 |
Long-range dependence in variable bit-rate video traffic
- Beran, Sherman, et al.
- 1995
(Show Context)
Citation Context ...ctable. The second dominant MPEG video traffic characteristic is the subexponential duration of scene lengths. Subexponential (long-range) dependence has been observed and studied in video traffic in =-=[10]-=-, where the modeling approach was through self-similar processes. Resnick and Samorodnitsky [11] investigated the long-range dependency of the autocorrelation function of video conference traces. We o... |

27 |
Tsiolis, “The impact of autocorrelation on queueing systems
- Livny, Melamed, et al.
- 1993
(Show Context)
Citation Context ...rival process ACF and the queue length distribution. It also explains why, in the recent literature, much attention has been given to investigating the impact of the ACF on the queueing behavior [3], =-=[31]-=-, [4]. The constant in the previous theorem can be explicitly calculated if the renewal distribution is assumed to be a regularly varying distribution . Recall from the definition of regular variation... |

22 | Multiplexing on-off sources with subexponential on periods
- Jelenković, Lazar
- 1997
(Show Context)
Citation Context ...dependency. A calculus of almost negligable computational complexity (unlike the case of self-similar processes) for the analysis of subexponential queueing systems was recently reported in [7], [1], =-=[13]-=-; in those papers, it is shown that queueing systems with subexponential arrival processes exhibit queueing behavior which is very distinct from the one obtained by exploring conventional exponential ... |

17 |
TES-based traffic modeling for performance evaluation of integrated networks
- Melamed, Raychaudhuri, et al.
- 1992
(Show Context)
Citation Context ...odeling video streams is to generate a stochastic process that matches its first- and second-order statistics, i.e., its marginal distribution and autocorrelation functions (ACF’s) (see, for example, =-=[22]-=-, [3], [4], [23]). For these methods to work well, it is desirable to have an easily computable expression of the ACF for the generated processJELENKOVIĆ et al.: MULTIPLE TIME SCALES AND SUBEXPONENTI... |

17 |
Modeling prioritized MPEG video using tes and a frame spreading strategy for transmission
- Ismail, Lambdaris, et al.
- 1995
(Show Context)
Citation Context ...reams is to generate a stochastic process that matches its first- and second-order statistics, i.e., its marginal distribution and autocorrelation functions (ACF’s) (see, for example, [22], [3], [4], =-=[23]-=-). For these methods to work well, it is desirable to have an easily computable expression of the ACF for the generated processJELENKOVIĆ et al.: MULTIPLE TIME SCALES AND SUBEXPONENTIALITY 1057 previ... |

17 | Fluid and diffusion limits for queues in slowly changing environments
- Choudhury, Mandelbaum, et al.
- 1996
(Show Context)
Citation Context ...ports the fluid models (see [29]) that have much lower computational complexity; 2) related asymptotic results in the Markovian framework that justify the fluid approximation were obtained in [9] and =-=[30]-=-; 3) this result can be proved under the weaker assumption that the distribution of the renewal times is intermediately regularly varying [13]. Due to the space constraints and the need for introducin... |

12 | Performance decay in a single server exponential queueing model with long range dependence
- Resnick, Samorodnitsky
- 1996
(Show Context)
Citation Context ...scene lengths. Subexponential (long-range) dependence has been observed and studied in video traffic in [10], where the modeling approach was through self-similar processes. Resnick and Samorodnitsky =-=[11]-=- investigated the long-range dependency of the autocorrelation function of video conference traces. We observed a subexponential scene length duration in MPEG video streams (this was also seen in [12]... |

12 | On the dependence of the queue tail distribution on multiple time scales of ATM multiplexers - Jelenković, Lazar - 1995 |

8 |
A theorem on sums of independent positive random variables and its application to branching random processes,” Theor
- Chistakov
- 1964
(Show Context)
Citation Context ...A distribution function on is called subexponential ( )if (12) where denotes the second convolution of with itself, i.e., . The class of subexponential distributions was first introduced by Chistakov =-=[32]-=-. Some examples of distribution functions in are the following. 1) Distributions of regular variations (contains Pareto distributions) 5) Benktander Type I distribution [34] and appropriately chosen. ... |

6 |
Tail probabilitites for a multiplexer with self-similar traffic
- Parulekar, Makowski
- 1996
(Show Context)
Citation Context ...e strict stability scenario, the dominant effect is due to the fast time scale buildups—put in the broader context some conflicting results from the literature. Indeed, a number of authors [3], [15], =-=[16]-=- contend that long-range dependence of one form or another has a dominant impact on the queue, while others claim that it does not [17], [18], [4]. Our framework allows one to synthesize these results... |

5 |
Sur un mode de croissance réguilière des fonctions,” Mathematica
- Karamata
- 1930
(Show Context)
Citation Context ...d at the renewal times, we denoted .) Lower Bound: First we will prove that . Since , we have , and being a slowly varying function, i.e., for all (this class of distributions was first introduced in =-=[33]-=-). 2) The lognormal distribution where is the standard normal distribution. 3) Weibull distribution From the inequality above, we see that the lower bound will follow if we manage to prove that . The ... |

4 |
Multiple time scale and subexponential asymptotic behavior of a network multiplexer
- Jelenković, Lazar
- 1996
(Show Context)
Citation Context ...ing analysis and for computationally inexpensive traffic generation. A distinct queueing behavior that results from the multiple time scale structure of arrival processes was recently investigated in =-=[7]-=- and [8]. In these papers, it was analytically shown that when a stream with multiple time scales passes through a queue, the queue length distribution has multiple decay rates. Similar results were i... |

4 |
A histogram based model for video traffic behavior in an ATM network node with an application to congestion control
- Skelly, Dixit, et al.
(Show Context)
Citation Context ...ults from the literature. Indeed, a number of authors [3], [15], [16] contend that long-range dependence of one form or another has a dominant impact on the queue, while others claim that it does not =-=[17]-=-, [18], [4]. Our framework allows one to synthesize these results1054 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 15, NO. 6, AUGUST 1997 in that the former can be viewed as instances of th... |

3 | Multiplexing ATM traffic streams with time-scale-dependent arrival processes,” preprint
- Landry, Stavrakakis
- 1995
(Show Context)
Citation Context ...or fractal models. These models essentially attempt to capture an infinite number of time scales and, for that reason, they generally suffer from high computational complexity. Landry and Stavrakakis =-=[6]-=- have presented a modeling approach based on multiple time scales that is appropriate for 0733–8716/97$10.00 © 1997 IEEEJELENKOVIĆ et al.: MULTIPLE TIME SCALES AND SUBEXPONENTIALITY 1053 cell and sli... |

3 |
Multiple timescales and subexponentiality in MPEG video streams
- Jelenković, Lazar, et al.
- 1996
(Show Context)
Citation Context ...namics. This process may be tedious, complex, and often involves a certain amount of “eyeballing.” 1 Preliminary modeling results that concentrate only on the multiple time structure were reported in =-=[14]-=-. In order to alleviate this problem, we also propose a direct method for extracting some statistical parameters of the video model that exhibits a very low computational complexity and can exactly ma... |

2 | A Markov-renewal-modulated TES model for VBR Star Wars video,” preprint
- Melamed, Pendarakis
(Show Context)
Citation Context ... and , with the 856 smallest scenes in the first and the other 198 scenes in the second. More sophisticated approaches to clustering scenes have been investigated in the literature (see, for example, =-=[20]-=-), but we find that they do not add fundamental insights to the main conclusions of this paper. Regime instances are formed by merging consecutive scenes of the same type. The average GOP sizes for th... |