## Analysis, Modeling and Generation of Self-Similar VBR Video Traffic (1994)

Citations: | 464 - 5 self |

### BibTeX

@INPROCEEDINGS{Garrett94analysis,modeling,

author = {Mark Garrett and Walter Willinger},

title = {Analysis, Modeling and Generation of Self-Similar VBR Video Traffic},

booktitle = {},

year = {1994},

pages = {269--280}

}

### Years of Citing Articles

### OpenURL

### Abstract

We present a detailed statistical analysis of a 2-hour long empirical sample of VBR video. The sample was obtained by applying a simple intraframe video compression code to an action movie. The main findings of our analysis are (1) the tail behavior of the marginal bandwidth distribution can be accurately described using "heavy-tailed" distributions (e.g., Pareto); (2) the autocorrelation of the VBR video sequence decays hyperbolically (equivalent to long-range dependence) and can be modeled using self-similar processes. We combine our findings in a new (non-Markovian) source model for VBR video and present an algorithm for generating synthetic traffic. Trace-driven simulations show that statistical multiplexing results in significant bandwidth efficiency even when long-range dependence is present. Simulations of our source model show long-range dependence and heavy-tailed marginals to be important components which are not accounted for in currently used VBR video traffic models. 1 I...

### Citations

1765 | On The Self-Similar Nature of Ethernet Traffic
- Leland, Taqqu, et al.
- 1994
(Show Context)
Citation Context ...ene activity. For video conferencing, for example, H tends to be smaller, typically between 0.60 - 0.75 [BERA93]. Computer traffic can be much more active, with measured H-values often close to unity =-=[LELA93]-=-. 4 VBR Video Model Construction and Traffic Generation We can now construct a source model for use in computer simulations that captures two important aspects from our analysis of the VBR video trace... |

1678 |
The fractal geometry of nature
- Mandelbrot
- 1983
(Show Context)
Citation Context ...rise to the term fractional noise processes. Similarly, the term fractal is used in the context of self-similar geometric objects which can be characterized using the notion of a fractional dimension =-=[MAND83]-=-. Hosking provides an algorithm for generating a long-range dependent process called fractional ARIMA(0; d; 0), where the zeros indicate there are no autoregressive (AR) and moving average (MA) parame... |

1246 | Simulation Modeling and Analysis - Law, Kelton |

870 | The JPEG still picture compression standard
- Wallace
- 1991
(Show Context)
Citation Context ...T) is computed. The DCT coefficients are uniformly quantized into 8 bits and compressed using run-length and Huffman coding. These algorithms comprise essentially the same coding as the JPEG standard =-=[WALL91]-=-. The quality of our coding is reasonable, except that block boundaries are noticeable in some cases. We assume that the fine tuning necessary for excellent visual quality would not change the charact... |

525 | Supporting RealTime Applications in an Integrated Services Packet Network: Architecture and Mechanisms
- Clark, Shenker, et al.
- 1992
(Show Context)
Citation Context ... Short periods of congestion may be handled gracefully through the use of prioritization(layered coding) [PVW91], congestion notification [GARR93], and intelligent scheduling algorithms at the switch =-=[CLAR92]-=-. However, good design and analysis of a network requires an understanding of the traffic itself. This paper presents a long trace of VBR video with statistical analyses and simulation results which a... |

494 | Fractional Brownian motions, fractional noises and applications - Mandelbrot, Ness - 1968 |

92 | Modeling persistence in hydrological time series using fractional differencing", Water Resources Research - Hosking - 1984 |

90 | Long-range dependence: A review
- Cox
- 1984
(Show Context)
Citation Context ...n one parameter to describe it!) Thus, the "ideal" LRD process (the kind that comes out of LRD models having H as the only timecorrelation parameter), is a (second-order) exactly self-simila=-=r process [COX84]. A covari-=-ance stationary process X is said to be (second-order) exactly self-similar if the corresponding "aggregated" processes X (m) have the same autocorrelation function as X, for all ms1, where ... |

66 |
Computer Experiments with Fractional Gaussian Noises
- Mandelbrot, Wallis
- 1969
(Show Context)
Citation Context ...ata (see Fig. 1). An LRD process plotted on a graph of any time scale, appears to have a dominant periodicity with a few cycles fitting on the plot. If more data is plotted, new dominant modes appear =-=[MAND69a]-=-. The increase of apparent amplitude withincreased time scale can be understood intuitively for video, especially movies. Within each scene there is random movement and variation of bandwidth. Changes... |

64 |
Contributions toward real-time services on packet switched networks
- Garrett
- 1993
(Show Context)
Citation Context ...ger dependence on motion result from interframe coding, i.e., coding frame differences or use of motion prediction /compensation. Our main results do seem to extend to interframe (MPEG) video as well =-=[GARR93a]-=-. (See also [PANC94] for analysis of an MPEG VBR video trace.) The next section describes the coding method and some interesting characteristics of the time series. Section 3 describes the basic stati... |

55 |
Robust R,/S analysis of long-run serial correlation
- Mandelbrot, Taqqu
- 1979
(Show Context)
Citation Context ... H-values of around 0.7. Others have since shown that (2) holds for short-range dependent processes with H = 0:5 [FELL51, MAND68], and for increment processes of selfsimilar models, where 0:5 ! H ! 1 =-=[MAND79]-=-. When working with empirical data, a practical implementation of the R=S analysis has been proposed by [MAND69a] (see also [MAND79]) and consists of plotting R(n)=S(n) versus n, for different lags n ... |

47 | Joint Source/Channel Coding of Statistically Multiplexed Real-Time Services on Packet Networks
- Garrett, Vetterli
- 1993
(Show Context)
Citation Context ...ectly allocate and regulate bandwidth assigned to the service. Short periods of congestion may be handled gracefully through the use of prioritization(layered coding) [PVW91], congestion notification =-=[GARR93]-=-, and intelligent scheduling algorithms at the switch [CLAR92]. However, good design and analysis of a network requires an understanding of the traffic itself. This paper presents a long trace of VBR ... |

43 |
R.: Some long run properties of geophysical records
- Mandelbrot, Wallis
- 1969
(Show Context)
Citation Context ...:5. Another graphical method for estimating the Hurst parameter from a given empirical record is called the "R=S analysis ". This method has been used to model a wide variety of geophysical =-=phenomena [MAND69b]-=- and is based on the log10(m) 0 1 2 3 4 Figure 11: Variance-time plot for VBR video trace. rescaled adjusted range statistics R=S, originally introduced by H. E. Hurst [HURS51]. The heuristic behind t... |

39 | The asymptotic distribution of the range of sums of independent random variables - Feller - 1951 |

23 |
The Transition and Autocorrelation
- Jagerman, Melamed
- 1992
(Show Context)
Citation Context ...e inverse cumulative probability function of our Gamma/Pareto model. (A similar technique for distorting the marginals is used where the original process is distributed Uniformly rather than Normally =-=[JAGE92]-=-.) F \Gamma=P requires only three parameters, and is constructed as follows: The Gamma distribution has the probability density function, f \Gamma (x) = e \Gammax (x) s\Gamma1 \Gamma(s) : (14) The sha... |

17 |
Zarki, "MPEG Coding For Variable Bit Rate Video Transmission
- Pancha, El
- 1994
(Show Context)
Citation Context ...ion result from interframe coding, i.e., coding frame differences or use of motion prediction /compensation. Our main results do seem to extend to interframe (MPEG) video as well [GARR93a]. (See also =-=[PANC94]-=- for analysis of an MPEG VBR video trace.) The next section describes the coding method and some interesting characteristics of the time series. Section 3 describes the basic statistics of the trace s... |

15 |
Variable bit rate video traffic and long range dependence
- Sherman, Taqqu, et al.
- 1994
(Show Context)
Citation Context ...ng evidence of long-range dependence (LRD), which is not accounted for in any of the commonly used stochastic models for VBR video traffic, and yet has been found to be ubiquitous in VBR video traces =-=[BERA93]-=-, and may have important effects on performance. See [GARR93a] for an extensive literature survey of VBR video statistical measurements and models. 2000 4000 6000 8000 10000 0.2 0.4 0.6 0.8 1 Time Lag... |

7 | Toward Joint Optimization of VBR Video Coding and Packet Network Tra c Control
- Ortega, Garrett, et al.
- 1993
(Show Context)
Citation Context ...d to accommodate the constant bit rate (CBR) channel of circuit-switched networks. Forcing the transmission rate to be constant results in delay, wasted bandwidth, and modulation of the video quality =-=[ORTE93]-=-. The availability of packet networks raises the question of whether real-time services (especially video due to its high bandwidth) can be improved in both efficiency and quality through variable rat... |

4 |
Long-Term Storage Capacity of Reservoirs", Trans
- Hurst
- 1951
(Show Context)
Citation Context ...r each object defined on a larger time scale, there is a larger variation of behavior. LRD is quantified by a single parameter, H , after H. E. Hurst who studied long-term storage in water reservoirs =-=[HURS51]-=-. H is related to the rate of decay fi of the autocorrelation coefficients and, equivalently, to the parameter ff that characterizes the power law behavior of the spectral density around the origin. W... |