1 Giuliano Casale, Eddy Z. Zhang, Evgenia Smirni {casale, eddy, Speaker: Giuliano Casale Numerical Methods for Structured Markov Chains,

Slides:



Advertisements
Similar presentations
Doc.: IEEE /1216r1 Submission November 2009 BroadcomSlide 1 Internet Traffic Modeling Date: Authors: NameAffiliationsAddressPhone .
Advertisements

A Large-Grained Parallel Algorithm for Nonlinear Eigenvalue Problems Using Complex Contour Integration Takeshi Amako, Yusaku Yamamoto and Shao-Liang Zhang.
Sampling plans for linear regression
Introduction Background Knowledge Workload Modeling Related Work Conclusion & Future Work Modeling Many-Task Computing Workloads on a Petaflop Blue Gene.
FTP Biostatistics II Model parameter estimations: Confronting models with measurements.
Regression Analysis Once a linear relationship is defined, the independent variable can be used to forecast the dependent variable. Y ^ = bo + bX bo is.
TNO orbit computation: analysing the observed population Jenni Virtanen Observatory, University of Helsinki Workshop on Transneptunian objects - Dynamical.
Data mining and statistical learning - lecture 6
DEPARTMENT OF HEALTH SCIENCE AND TECHNOLOGY STOCHASTIC SIGNALS AND PROCESSES Lecture 1 WELCOME.
September 2008CORTONA-ITALY, DOUBLY STRUCTURED SETS OF SYMPLECTIC MATRICES Froilán M. Dopico Instituto de Ciencias Matemáticas CSIC-UAM-UC3M-UCM.
CMPT 855Module Network Traffic Self-Similarity Carey Williamson Department of Computer Science University of Saskatchewan.
Motion Analysis (contd.) Slides are from RPI Registration Class.
Statistics & Modeling By Yan Gao. Terms of measured data Terms used in describing data –For example: “mean of a dataset” –An objectively measurable quantity.
A gentle introduction to fluid and diffusion limits for queues Presented by: Varun Gupta April 12, 2006.
A Nonstationary Poisson View of Internet Traffic T. Karagiannis, M. Molle, M. Faloutsos University of California, Riverside A. Broido University of California,
1 KPC-Toolbox Demonstration Eddy Zheng Zhang, Giuliano Casale, Evgenia Smirni Computer Science Department College of William & Mary.
A First Peek at the Extremogram: a Correlogram of Extremes 1. Introduction The Autocorrelation function (ACF) is widely used as a tool for measuring Serial.
01/22/09ICDCS20061 Load Unbalancing to Improve Performance under Autocorrelated Traffic Ningfang Mi College of William and Mary Joint work with Qi Zhang.
AN ALGORITHM FOR TESTING UNIDIMENSIONALITY AND CLUSTERING ITEMS IN RASCH MEASUREMENT Rudolf Debelak & Martin Arendasy.
Testing Bridge Lengths The Gadsden Group. Goals and Objectives Collect and express data in the form of tables and graphs Look for patterns to make predictions.
Computer Science Characterizing and Exploiting Reference Locality in Data Stream Applications Feifei Li, Ching Chang, George Kollios, Azer Bestavros Computer.
Correctness of Gossip-Based Membership under Message Loss Maxim Gurevich, Idit Keidar Technion.
Relationships Among Variables
EE513 Audio Signals and Systems Statistical Pattern Classification Kevin D. Donohue Electrical and Computer Engineering University of Kentucky.
Self-Similarity of Network Traffic Presented by Wei Lu Supervised by Niclas Meier 05/
Panel Topic: After Long Range Dependency (LRD) discoveries, what are the lessons learned so far to provide QoS for Internet advanced applications David.
References for M/G/1 Input Process
Statistical Methods For Engineers ChE 477 (UO Lab) Larry Baxter & Stan Harding Brigham Young University.
Stability Analysis of Linear Switched Systems: An Optimal Control Approach 1 Michael Margaliot School of Elec. Eng. Tel Aviv University, Israel Joint work.
Relationship of two variables
1 MULTI VARIATE VARIABLE n-th OBJECT m-th VARIABLE.
Simple Linear Regression
Isolated-Word Speech Recognition Using Hidden Markov Models
Alignment Introduction Notes courtesy of Funk et al., SIGGRAPH 2004.
Oceanography 569 Oceanographic Data Analysis Laboratory Kathie Kelly Applied Physics Laboratory 515 Ben Hall IR Bldg class web site: faculty.washington.edu/kellyapl/classes/ocean569_.
 1  Outline  stages and topics in simulation  generation of random variates.
Yoni Nazarathy Gideon Weiss University of Haifa Yoni Nazarathy Gideon Weiss University of Haifa The Asymptotic Variance of the Output Process of Finite.
© 2009 IBM Corporation 1 Improving Consolidation of Virtual Machines with Risk-aware Bandwidth Oversubscription in Compute Clouds Amir Epstein Joint work.
Texture. Texture is an innate property of all surfaces (clouds, trees, bricks, hair etc…). It refers to visual patterns of homogeneity and does not result.
Educational Research: Competencies for Analysis and Application, 9 th edition. Gay, Mills, & Airasian © 2009 Pearson Education, Inc. All rights reserved.
Time Series Data Analysis - I Yaji Sripada. Dept. of Computing Science, University of Aberdeen2 In this lecture you learn What are Time Series? How to.
1 Multivariate Linear Regression Models Shyh-Kang Jeng Department of Electrical Engineering/ Graduate Institute of Communication/ Graduate Institute of.
ENM 503 Lesson 1 – Methods and Models The why’s, how’s, and what’s of mathematical modeling A model is a representation in mathematical terms of some real.
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Binary Stochastic Fields: Theory and Application to Modeling of Two-Phase Random Media Steve Koutsourelakis University of Innsbruck George Deodatis Columbia.
ICOM 6115: Computer Systems Performance Measurement and Evaluation August 11, 2006.
Yoni Nazarathy Gideon Weiss University of Haifa Yoni Nazarathy Gideon Weiss University of Haifa The Asymptotic Variance Rate of the Departure Process of.
Mathematical Models & Optimization?
© 2006 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice Injecting Realistic Burstiness to.
Optimal Sampling Strategies for Multiscale Stochastic Processes Vinay Ribeiro Rolf Riedi, Rich Baraniuk (Rice University)
Learning to Detect Events with Markov-Modulated Poisson Processes Ihler, Hutchins and Smyth (2007)
1/18 New Feature Presentation of Transition Probability Matrix for Image Tampering Detection Luyi Chen 1 Shilin Wang 2 Shenghong Li 1 Jianhua Li 1 1 Department.
Accurate Robot Positioning using Corrective Learning Ram Subramanian ECE 539 Course Project Fall 2003.
Correlation & Regression Analysis
OPERATING SYSTEMS CS 3530 Summer 2014 Systems and Models Chapter 03.
Chapter 15 The Chi-Square Statistic: Tests for Goodness of Fit and Independence PowerPoint Lecture Slides Essentials of Statistics for the Behavioral.
Chapter 2-OPTIMIZATION G.Anuradha. Contents Derivative-based Optimization –Descent Methods –The Method of Steepest Descent –Classical Newton’s Method.
A Kernel Approach for Learning From Almost Orthogonal Pattern * CIS 525 Class Presentation Professor: Slobodan Vucetic Presenter: Yilian Qin * B. Scholkopf.
7-1 MGMG 522 : Session #7 Serial Correlation (Ch. 9)
I. Statistical Methods for Genome-Enabled Prediction of Complex Traits OUTLINE THE CHALLENGES OF PREDICTING COMPLEX TRAITS ORDINARY LEAST SQUARES (OLS)
Biointelligence Laboratory, Seoul National University
Bounded Nonlinear Optimization to Fit a Model of Acoustic Foams
OPERATING SYSTEMS CS 3502 Fall 2017
STATISTICAL ORBIT DETERMINATION Kalman (sequential) filter
Perturbation method, lexicographic method
Accurate Robot Positioning using Corrective Learning
Random walk initialization for training very deep feedforward networks
Tabulations and Statistics
Feifei Li, Ching Chang, George Kollios, Azer Bestavros
CPSC 641: Network Traffic Self-Similarity
Presentation transcript:

1 Giuliano Casale, Eddy Z. Zhang, Evgenia Smirni {casale, eddy, Speaker: Giuliano Casale Numerical Methods for Structured Markov Chains, Dagstuhl Seminar November 11-14, 2007 College of William & Mary Department of Computer Science Williamsburg, , Virginia, US Interarrival Times Characterization and Fitting for Markovian Traffic Analysis

2 Outline Motivations Motivations Review of MAP Fitting Algorithms Review of MAP Fitting Algorithms from fitting counts to interarrival times (IAT) fitting from fitting counts to interarrival times (IAT) fitting observations on eigenvalue-based methods observations on eigenvalue-based methods Jordan characterization of MAP moments and autocorrelations Jordan characterization of MAP moments and autocorrelations analysis of small MAPs analysis of small MAPs Composition of large MAPs Composition of large MAPs MAP fitting using higher-order correlations MAP fitting using higher-order correlations

3 Motivation MAP/MMPP Model Parameterization MAP/MMPP Model Parameterization Markovian models of network traffic Markovian models of network traffic MAP closed queueing networks (see slides E. Smirni) ‏ MAP closed queueing networks (see slides E. Smirni) ‏ MAP fitting is not fully understood MAP fitting is not fully understood E.g., some questions: E.g., some questions: Fit the counting process or the interarrival process? Fit the counting process or the interarrival process? How many moments? Which correlation coeffs? How many moments? Which correlation coeffs? How fitting decisions affect queueing prediction? How fitting decisions affect queueing prediction? Is nonlinear optimization appropriate? Is nonlinear optimization appropriate?

4 MMPP Counting Process Fitting Measuring counts in networks can be often easier than measuring interarrival times Measuring counts in networks can be often easier than measuring interarrival times S. Li & C.L. Hwang, 1992, 1993: S. Li & C.L. Hwang, 1992, 1993: Circulant matrices to impose MMPP power spectrum Circulant matrices to impose MMPP power spectrum A.T. Andersen & B.F. Nielsen, 1998: A.T. Andersen & B.F. Nielsen, 1998: Superposition of MMPP(2)s (Kronecker sum) ‏ Superposition of MMPP(2)s (Kronecker sum) ‏ Matching of the Hurst parameter Matching of the Hurst parameter Degrees of freedom for optional least-square fitting of the interarrival time (IAT) autocorrelations (ACF) ‏ Degrees of freedom for optional least-square fitting of the interarrival time (IAT) autocorrelations (ACF) ‏ Good accuracy on the Bellcore Aug89/Oct89 traces Good accuracy on the Bellcore Aug89/Oct89 traces

5 MAP Counting Process Fitting A. Horváth & M. Telek, 2002: A. Horváth & M. Telek, 2002: Multifractal traffic model, e.g., Riedi et al., 1999 Multifractal traffic model, e.g., Riedi et al., 1999 Traffic analysis based on Haar wavelet transform Traffic analysis based on Haar wavelet transform Each MAP(2) describes variability in the Haar wavelet coefficients at a specific time scale Each MAP(2) describes variability in the Haar wavelet coefficients at a specific time scale Almost optimal fitting of the BC-Aug89 trace Almost optimal fitting of the BC-Aug89 trace Further improvements may not be easy: Further improvements may not be easy: Higher-order moments of counts hard to manipulate Higher-order moments of counts hard to manipulate

6 MAP Interarrival Process Two-phase fitting fitting of PH-type distribution followed by fitting of IAT ACF Two-phase fitting fitting of PH-type distribution followed by fitting of IAT ACF Feasible manipulation of higher-order moments Feasible manipulation of higher-order moments P. Buchholz et al., 2003, 2004: P. Buchholz et al., 2003, 2004: Expectation Maximization (EM) algorithms Expectation Maximization (EM) algorithms Support for two-phase fitting Support for two-phase fitting Scalability of EM rapidly increasing (Panchenko & Thümmler, 2007) ‏ Scalability of EM rapidly increasing (Panchenko & Thümmler, 2007) ‏

7 MAP Interarrival Process Moment and ACF Analytical Fitting: Moment and ACF Analytical Fitting: Results only for MMPP(2), MAP(2), MAP(3) ‏ Results only for MMPP(2), MAP(2), MAP(3) ‏ G. Horváth, M. Telek & P. Buchholz, 2005: G. Horváth, M. Telek & P. Buchholz, 2005: Two-phase least-square fitting of PH distrib. and ACF Two-phase least-square fitting of PH distrib. and ACF Optimization variables are the MAP transition rates, i.e., the O(n 2 ) entries of the D 0 and D 1 matrices Optimization variables are the MAP transition rates, i.e., the O(n 2 ) entries of the D 0 and D 1 matrices Simple to understand and implement Simple to understand and implement Least-squares can be numerically difficult: Least-squares can be numerically difficult: small magnitude of transition rates compared to tolerance small magnitude of transition rates compared to tolerance infeasibility due to inappropriate choice of step size infeasibility due to inappropriate choice of step size

8 Our observations Observation 1: eigenvalues give direct control to the nonlinear solver on ACF decay and CDF tail Observation 1: eigenvalues give direct control to the nonlinear solver on ACF decay and CDF tail Observation 2: lack of general Jordan analysis of IAT moments and autocorrelations Observation 2: lack of general Jordan analysis of IAT moments and autocorrelations Observation 3: eigenvalue-based least-squares tends to be numerically well-behaved Observation 3: eigenvalue-based least-squares tends to be numerically well-behaved Observation 4: inverse eigenvalue problems often prohibitive, how do we determine D 0 and D 1 ? Observation 4: inverse eigenvalue problems often prohibitive, how do we determine D 0 and D 1 ?  Superposition does not help for IAT process

9 Our contributions A general Jordan analysis of MAP moments and autocorrelations A general Jordan analysis of MAP moments and autocorrelations Using this characterization we analyze the IAT process in small MAPs Using this characterization we analyze the IAT process in small MAPs We find a compositional approach to define the IAT process in large MAPs using small MAPs We find a compositional approach to define the IAT process in large MAPs using small MAPs Main result: A least-squares that can fit IAT moments and correlations of any order Main result: A least-squares that can fit IAT moments and correlations of any order

10 Why statistics of “any order”? Literature evaluates up to second-order properties Literature evaluates up to second-order properties Higher-order correlations are neglected, but.... Higher-order correlations are neglected, but....

11 MAP Jordan Analysis Definition: MAP moments Definition: MAP moments Definition: MAP autocorrelations Definition: MAP autocorrelations Moments and correlations depend on matrix powers Moments and correlations depend on matrix powers Eigenvalues explicited by the Cayley-Hamilton theorem Eigenvalues explicited by the Cayley-Hamilton theorem

12 MAP Jordan Analysis – Cont'd

13 MAP(3) Characterization Example Define MAPs with given oscillatory ACF Define MAPs with given oscillatory ACF Generalization of Circulant MAPs to IAT process Generalization of Circulant MAPs to IAT process MAP Definition Characterization

14 MAP(3) Characterization Example SCV=4.87 p1=0 p2=0.0286

15 Composition of Large Processes Idea: use Kronecker product to overcome inverse eigenvalue problem in eigenvalue-based fittings Idea: use Kronecker product to overcome inverse eigenvalue problem in eigenvalue-based fittings Kronecker product composition (KPC) Kronecker product composition (KPC) One of the two D 0 matrices must be diagonal One of the two D 0 matrices must be diagonal No loss of generality No loss of generality Prevents negative (infeasible) off-diagonal entries in the D 0 matrix of the KPC Prevents negative (infeasible) off-diagonal entries in the D 0 matrix of the KPC

16 Jordan Analysis of KPC process Eigenvalues and projectors Eigenvalues and projectors Moments and autocorrelations Moments and autocorrelations

17 KPC Example To the best of our knowledge, never shown in the literature a MAP with lag-1 acf > 0.5 To the best of our knowledge, never shown in the literature a MAP with lag-1 acf > 0.5 Does it exist? Does it exist? MAP(2) must have lag-1 acf <0.5 MAP(2) must have lag-1 acf <0.5 Not found in random MAP(3) and MAP(4) Not found in random MAP(3) and MAP(4) Answer: yes it exists, it can be defined by KPC Answer: yes it exists, it can be defined by KPC A simple MAP(2): A simple MAP(2):

18 KPC Example – Cont'd What happens if we compose with KPC the MAP(2) with a PH renewal process? What happens if we compose with KPC the MAP(2) with a PH renewal process? Composition with a hypoexponential process Composition with a hypoexponential process

19 Jordan Analysis of KPC process IAT Joint Moments Joint moments, e.g.,G. Horváth & M. Telek, 2007 Joint moments, e.g.,G. Horváth & M. Telek, 2007 Admits characterization similar to moments/acf Admits characterization similar to moments/acf Joint moments in KPC process Joint moments in KPC process Conclusion: KPC can fit moments of any order Conclusion: KPC can fit moments of any order

20 Two-Phase Least Squares We determine J small MAPs to be composed by KPC in order to best fit a trace We determine J small MAPs to be composed by KPC in order to best fit a trace Lessons learned from Jordan analysis: Lessons learned from Jordan analysis: first fit ACF and SCV, then moments first fit ACF and SCV, then moments Phase 2: Fit moments Phase 1: ACV+SCV eigenvalue-based mean and bispectrum

21 Results: BC-Aug89 quality of fitting - MAP(16) ‏

22 Results: Seagate-Web quality of fitting - MAP(16) ‏

23 Results: BC-Aug89 queueing prediction - MAP(16) ‏

24 Results: Seagate Web queueing prediction - MAP(16) ‏

25 Conclusion Jordan characterization allows: Jordan characterization allows: analysis of simple MAP processes analysis of simple MAP processes least-square fitting that is numerically well-behaved least-square fitting that is numerically well-behaved Joint IAT moments required for accurate queueing prediction of real workloads Joint IAT moments required for accurate queueing prediction of real workloads even bispectrum fitting leaves room for improvement even bispectrum fitting leaves room for improvement KPC indispensable for definition of large processes KPC indispensable for definition of large processes Future work Future work Fitting traces with strong oscillatory patterns (e.g., MPEG traces) ‏ Fitting traces with strong oscillatory patterns (e.g., MPEG traces) ‏ Comparison with circulant MAPs approach Comparison with circulant MAPs approach