Multiscale Models for Network Traffic Vinay Ribeiro Rolf Riedi, Matt Crouse, Rich Baraniuk Dept. of Electrical Engineering Rice University (Houston, Texas)

Slides:



Advertisements
Similar presentations
pathChirp Efficient Available Bandwidth Estimation
Advertisements

Min-Plus Linear Systems Theory and Bandwidth Estimation Min-Plus Linear Systems Theory and Bandwidth Estimation TexPoint fonts used in EMF. Read the TexPoint.
Principal Component Analysis Based on L1-Norm Maximization Nojun Kwak IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008.
Chapter 3. Noise Husheng Li The University of Tennessee.
Fast Algorithms For Hierarchical Range Histogram Constructions
Applications in Signal and Image Processing
Exploiting Sparse Markov and Covariance Structure in Multiresolution Models Presenter: Zhe Chen ECE / CMR Tennessee Technological University October 22,
Multiscale Analysis for Intensity and Density Estimation Rebecca Willett’s MS Defense Thanks to Rob Nowak, Mike Orchard, Don Johnson, and Rich Baraniuk.
2008 SIAM Conference on Imaging Science July 7, 2008 Jason A. Palmer
STAT 497 APPLIED TIME SERIES ANALYSIS
2  Something “feels the same” regardless of scale 4 What is that???
Wavelet-Based Network Traffic Modeling Carey Williamson University of Calgary.
CMPT 855Module Network Traffic Self-Similarity Carey Williamson Department of Computer Science University of Saskatchewan.
On the Self-Similar Nature of Ethernet Traffic - Leland, et. Al Presented by Sumitra Ganesh.
Wavelet Estimation of a Local Long Memory Parameter B. Whitcher EURANDOM, The Netherlands M. J. Jensen University of Missouri.
Introduction to Compressive Sensing
Parametric Inference.
A Nonstationary Poisson View of Internet Traffic T. Karagiannis, M. Molle, M. Faloutsos University of California, Riverside A. Broido University of California,
1 Using A Multiscale Approach to Characterize Workload Dynamics Characterize Workload Dynamics Tao Li June 4, 2005 Dept. of Electrical.
Rethinking Internet Traffic Management: From Multiple Decompositions to a Practical Protocol Jiayue He Princeton University Joint work with Martin Suchara,
Wavelet-based Coding And its application in JPEG2000 Monia Ghobadi CSC561 project
Self-Similar through High-Variability: Statistical Analysis of Ethernet LAN Traffic at the Source Level Walter Willinger, Murad S. Taqqu, Robert Sherman,
Statistical Methods for long-range forecast By Syunji Takahashi Climate Prediction Division JMA.
Lecture II-2: Probability Review
Review of Probability.
Internet Traffic Modeling Poisson Model vs. Self-Similar Model By Srividhya Chandrasekaran Dept of CS University of Houston.
Informational Network Traffic Model Based On Fractional Calculus and Constructive Analysis Vladimir Zaborovsky, Technical University, Robotics Institute,
INDEPENDENT COMPONENT ANALYSIS OF TEXTURES based on the article R.Manduchi, J. Portilla, ICA of Textures, The Proc. of the 7 th IEEE Int. Conf. On Comp.
Heart Sound Background Noise Removal Haim Appleboim Biomedical Seminar February 2007.
1 FARIMA(p,d,q) Model and Application n FARIMA Models -- fractional autoregressive integrated moving average n Generating FARIMA Processes n Traffic Modeling.
Multiscale Traffic Processing Techniques for Network Inference and Control Richard Baraniuk Edward Knightly Robert Nowak Rolf Riedi Rice University INCITE.
Direct Message Passing for Hybrid Bayesian Networks Wei Sun, PhD Assistant Research Professor SFL, C4I Center, SEOR Dept. George Mason University, 2009.
Basics of Neural Networks Neural Network Topologies.
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: Definitions Random Signal Analysis (Review) Discrete Random Signals Random.
Link Dimensioning for Fractional Brownian Input Chen Jiongze PhD student, Electronic Engineering Department, City University of Hong Kong Supported by.
Digital Image Processing CSC331 Image Enhancement 1.
ELEC 303 – Random Signals Lecture 18 – Classical Statistical Inference, Dr. Farinaz Koushanfar ECE Dept., Rice University Nov 4, 2010.
Elements of Stochastic Processes Lecture II
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: ML and Simple Regression Bias of the ML Estimate Variance of the ML Estimate.
CHEE825 Fall 2005J. McLellan1 Spectral Analysis and Input Signal Design.
Image Denoising Using Wavelets
Multiscale Network Processes: Fractal and p-Adic analysis Vladimir Zaborovsky, Technical University, Robotics Institute, Saint-Petersburg, Russia .
Multiplicative Wavelet Traffic Model and pathChirp: Efficient Available Bandwidth Estimation Vinay Ribeiro.
Introduction to Digital Signals
University of Houston Cullen College of Engineering Electrical & Computer Engineering Capacity Scaling in MIMO Wireless System Under Correlated Fading.
Optimal Sampling Strategies for Multiscale Stochastic Processes Vinay Ribeiro Rolf Riedi, Rich Baraniuk (Rice University)
PathChirp Spatio-Temporal Available Bandwidth Estimation Vinay Ribeiro Rolf Riedi, Richard Baraniuk Rice University.
Lecture 2: Statistical learning primer for biologists
ENEE631 Digital Image Processing (Spring'04) Basics on 2-D Random Signal Spring ’04 Instructor: Min Wu ECE Department, Univ. of Maryland, College Park.
Rice University, SPiN Group spin.rice.edu Multiscale Traffic Processing Techniques for Network Inference and Control R. Baraniuk R. Nowak E. Knightly R.
Multiplicative Cascade Modeling of Computer Network Traffic Patricia H. Carter B10, NSWCCDD Interface 2002 April 19,2002.
D. Rincón, M. Roughan, W. Willinger – Towards a Meaningful MRA of Traffic Matrices 1/36 Towards a Meaningful MRA for Traffic Matrices D. Rincón, M. Roughan,
MODELING THE SELF-SIMILAR BEHAVIOR OF PACKETIZED MPEG-4 VIDEO USING WAVELET-BASED METHODS Dogu Arifler and Brian L. Evans The University of Texas at Austin.
Performance Evaluation of Long Range Dependent Queues Performance Evaluation of Long Range Dependent Queues Chen Jiongze Supervisor: Moshe ZukermanCo-Supervisor:
Risk Analysis Workshop April 14, 2004 HT, LRD and MF in teletraffic1 Heavy tails, long memory and multifractals in teletraffic modelling István Maricza.
Internet Measurement and Analysis Vinay Ribeiro Shriram Sarvotham Rolf Riedi Richard Baraniuk Rice University.
PathChirp & STAB Measuring Available Bandwidth and Locating Bottlenecks in Packet Networks Vinay Ribeiro Rolf Riedi, Richard Baraniuk Rice University spin.rice.edu.
PathChirp Efficient Available Bandwidth Estimation Vinay Ribeiro Rice University Rolf Riedi Rich Baraniuk.
Jun Li 1, Zhongdong Yang 1, W. Paul Menzel 2, and H.-L. Huang 1 1 Cooperative Institute for Meteorological Satellite Studies (CIMSS), UW-Madison 2 NOAA/NESDIS/ORA.
Optimization-based Cross-Layer Design in Networked Control Systems Jia Bai, Emeka P. Eyisi Yuan Xue and Xenofon D. Koutsoukos.
Tree and Forest Classification and Regression Tree Bagging of trees Boosting trees Random Forest.
Locating a Shift in the Mean of a Time Series Melvin J. Hinich Applied Research Laboratories University of Texas at Austin
Queuing Analysis of Tree-Based LRD Traffic Models Vinay J. Ribeiro R. Riedi, M. Crouse, R. Baraniuk.
Probability Theory and Parameter Estimation I
Fast Approximate Query Answering over Sensor Data with Deterministic Error Guarantees Chunbin Lin Joint with Etienne Boursier, Jacque Brito, Yannis Katsis,
Statistical Methods For Engineers
Presented by Chun Zhang 2/14/2003
Richard Baraniuk Edward Knightly Robert Nowak Rolf Riedi
pathChirp Efficient Available Bandwidth Estimation
pathChirp Efficient Available Bandwidth Estimation
Presentation transcript:

Multiscale Models for Network Traffic Vinay Ribeiro Rolf Riedi, Matt Crouse, Rich Baraniuk Dept. of Electrical Engineering Rice University (Houston, Texas)

Outline Multiscale nature of network traffic Wavelets Wavelet models for traffic Network inference applications

Time Scales time unit 2 n 2 1 (discrete time)

Multiscale Nature of Network Traffic Network traffic (local area networks, wide area networks, video traffic etc.) - variance decays slowly with aggregation i.i.d. data – variance decays faster with aggregation 60ms 6ms time unit 600ms Internet bytes/time trace (LBL’93) i.i.d. time series (lognormal)

Fractional Gaussian Noise (fGn) Stationary Gaussian process, Covariance (Hurst parameter: 0<H<1) Long-range dependence (LRD) if ½<H<1 Second-order self-similarity Variance-time plot

fGn is a 1/f-Process Power spectral density decays in a 1/f fashion –Low frequency components  long-term correlations frequency power

Towards Generalizations of fGn Variance decay of traffic not always straight line like fGn Goal: develop LRD models –Generalize fGn –Parsimonious (few parameters) –Fast synthesis for simulations Variance-time plot Auckland Univ. Traffic time scale

Wavelets Consider only orthonormal wavelet basis in L 2 (R) Prototype functions approximation function- wavelet function- Basis formed by scaled and shifted versions of prototype functions Approximation and wavelet coefficients

The Haar Wavelet Basis

Computing the Haar Transform Wavelet Transform: fine to coarse (bottom to top) Inverse Wavelet Transform: coarse to fine (top to bottom)

Wavelets and Filtering Wavelet coefficients at any scale j is the output of a bandpass filter Coarse scales  low frequency band Fine scales  high frequency band Width of bandpass filters increase exponentially frequency

Wavelets “Decorrelate” 1/f Processes Analysis of 1/f data –sample means converge faster in wavelet domain –estimate H in wavelet domain Synthesis of 1/f data –Exploit weak correlation in wavelet domain –Generate independent wavelet coefficients with appropriate variance –Invert wavelet transform frequency 1/f spectrum time domain 1/f strong correlation wavelet domain not 1/f weak correlation power

Haar Wavelet “Additive” 1/f Model Choose W j,k i.i.d. within scale j Set var( W j,k ) to obtain required decay of var( V j,k ) Fast O( N ) synthesis log 2 (N) parameters Asymptotically Gaussian

Sample Realization Realization is Gaussian and can take negative values Network traffic may be non-Gaussian and is always positive

Multiplicative Cascade Model Replace additive innovations by multiplicative innovations A j,k 2 [0,1], example -distribution Choose var( A j,k ) to get appropriate decay of var( V j,k ) Fast O(N) synthesis log 2 (N) parameters Positive data Asymptotically lognormal at fine time scales

Sample Realization Data is positive Same var(V j,k ) as additive model

Additive vs. Multiplicative Models Multiplicative model marginals closer to real data than additive model Additive model Multiplicative model Internet data (Auckland Univ) 6ms 12ms 24ms time unit

Queuing Experiment Additive and multiplicative models same var(V j,k ) Multiplicative model outperforms additive model High-order moments can influence queuing (open loop) real traffic multiplicative model additive model Kilo bytes

Shortcomings of Multiscale Models Open-loop –Do not capture closed-loop nature of network protocols and user behavior Physical intuition –Cascades model “redistribution” of traffic (multiplexing at queues, TCP)? Stationarity: first order stationary but not second- order stationary –Time averaged correlation structure is close to fGn –Queuing of additive model close to stationary Gaussian data (simulations and theory)

Selected References Self-similar traffic and networks (upto 1996) –W. Willinger, M. Taqqu, A. Erramilli, “A bibliographical guide to self-similar traffic and performance modeling for modern high-speed networks”, Stochastic Networks: Theory and Applications, vol. 4, Oxford Univ. Press, Wavelets –S. Burrus and R. Gopinath, “Introduction to Wavelets and Wavelet Transforms”, Prentice Hall, –I. Daubechies, “Ten lectures on wavelets”, SIAM, New York, Additive model –S. Ma and C. Ji, “Modeling heterogeneous network traffic in wavelet domain”, IEEE Trans. Networking, vol. 9, no. 5, Oct Multiplicative model –R. Riedi, M. Crouse, V. Ribeiro, R. Baraniuk, “A multifractal wavelet model with application to network traffic”, IEEE Trans. Info. Theory, vol. 45, no. 3, April –A. Feldmann, A. C. Gilbert, W. Willinger, “Data networks as cascasdes: investigating the multifractal nature of Internet WAN traffic”, ACM SIGCOMM, pp , –P. Mannersalo and I. Norros, “Multifractal analysis of real ATM traffic: A first look”, Technical report, VTT Information Technology, 1997, COST257TD(97)19,

Network Inference Applications

Why Network Inference? Each dot is one Internet Service Provider Different parts of Internet owned by different organizations Information sharing difficult –Commerical interests/trade secrets –Privacy –Sheer volume of “network state”

Edge-based Probing Inject probe packets into network Infer internal properties from packet delay/loss Current tools infer –Topology –Link bandwidths –End-to-end available bandwidth –Congestion locations

Cross-Traffic Inference Simple network path – single queue Spread of packet pair gives cross-traffic over small time interval 

Inferring cross-traffic over large time interval [0,T] Probing uncertainty principle –Dense sampling: accurate inference, affect cross- traffic –Sparse sampling: less accurate inference, less influence on cross-traffic

Problem Statement Given N probe pairs, how must we space them over time interval [0,T] to optimally estimate the total cross-traffic in [0,T] Answer depends on –cross-traffic –optimality criterion

Multiscale Cross-Traffic Model Choose N leaf nodes to give best linear estimate (in terms of mean squared error) of root node Take a guess! –Bunch probes together –Exponentially space probes pairs –Uniformly space probes over interval –Your favorite solution root leaves

Sensor Networks Application Each sensor samples local value of process (pollution, temperature etc.) Sensors cost money! Find best placement for N sensors to measure global average Global average possible sensor location

Independent Innovations Trees Each node is a linear combination of parent and an independent random innovation Optimal solution obtained by a water-filling procedure

: arbitrary set of leaf nodes; : size of X : leaves on left, : leaves on right : linear min. mean sq. error of estimating root using X Water-Filling f L (l)f R (l) N= Repeat at next lower scale with N replaced by l * N (left) and (N-l * N ) (right) Result: If innovations identically distributed within each scale then uniformly distribute leaves, l * N = b N/2 c

Covariance Trees Distance : Two leaf nodes have distance j if their lowest common ancestor is at scale j Covariance tree : Covariance between leaf nodes with distance j is c j (only a function of distance), covariance between root and any leaf node is constant,  Positively correlated tree : c j >c j+1 Negatively correlated tree : c j <c j+1

Covariance Tree Result Result: For a positively correlated tree choosing leaf nodes uniformly in the tree is optimal. However, for negatively correlated trees this same uniform choice is the worst case! Optimality proof: Simply construct an independent innovations tree with similar correlation structure Worst case proof: The uniform choice maximizes sum of elements of S X Using eigen analysis show that this implies that uniform choice minimizes sum of elements of S -1 X

Future Directions Sampling –More general tree structures –Non-linear estimates –Non-tree stochastic processes Traffic estimation –More complex networks Sensor networks – jointly optimize with other constraints like power transmission

References Estimation on multiscale trees –A. Willsky, “Multiresolution Markov models for signal and image processing”, Proc. of the IEEE 90(8), August Optimal sampling on trees –V. Ribeiro, R. Riedi, and R. Baraniuk, “Optimal sampling strategies for multiscale models and their application to computer networks”, preprint.