Browsing by Author "Mathur, Anup"

Now showing 1 - 6 of 6
  • Thumbnail Image
    Chitra: Visual Analysis of Parallel and Distributed Programs in the Time, Event, and Frequency Domains
    Abrams, Marc; Doraswamy, Naganand; Mathur, Anup (Department of Computer Science, Virginia Polytechnic Institute & State University, 1992)
    Chitra analyzes a program execution sequence (PES) collected during execution of a program and produces a homogeneous, semi-Markov chain model fitting the PES. The PES represents the evolution of a program state vector in time. Therefore Chitra analyzes the time-dependent behavior of a program. The paper describes a set of transforms that map a PES to a simplified PES. Because the transforms are program-independent, Chitra can be used with any program. Chitra provides a visualization of PES's and transforms, to allow a user visually to guide transform selection in an effort to generate a simple yet accurate semi-Markov chain model. The resultant chain can predict performance at program parameters different than those used in the input PES, and the chain structure can diagnose performance problems.
  • Modeling Transcient Trace Data
    Mathur, Anup; Abrams, Marc (Department of Computer Science, Virginia Polytechnic Institute & State University, 1996-10-01)
    This paper introduces a novel technique to construct an empirical workload model fitting time-varying (transient) trace data. The trace can be a categorical or numerical time-series. We model the trace as a Piecewise Independent stochastic process. To estimate the parameters for our model we first build a Rate Evolution Graph from the trace data. Piecewise linear regression is then used to construct a joint time-dependent probablity mass function for the trace data. Two methods are proposed to build a parsi- monious model. The modeling approach is demonstrated by the application of our model to twelve traces from the performance analysis domain.
  • Thumbnail Image
    A stochastic process model for transient trace data
    Mathur, Anup (Virginia Tech, 1996)
    Creation of sufficiently accurate workload models of computer systems is a key step in evaluating and tuning these systems. Workload models for an observable system can be built from traces collected by observing the system. This dissertation presents a novel technique to construct non-executable, artificial workload models fitting transient trace data. The trace can be a categorical or numerical time-series. The trace is considered a sample realization of a non-stationary stochastic process, {Xt}, such that random variables Xt follow different probability distributions. To estimate the parameters for the model a Rate Evolution Graph (REG) is built from the trace data. The REG is a two-dimensional Cartesian graph which plots the number of occurrences of each unique state in the trace on the ordinate and time on the abscissa. The REG contains one path for all instances of each unique state in the trace. The derivative of a REG path at time t is used as an estimate of the probability of occurrence of the corresponding state at t. We use piecewise linear regression to fit straight line segments to each REG path. The slopes of the line segments that fit a REG path estimate the time dependent probability of occurrence of the corresponding state. The estimates of occurrence probabilities of all unique states in the trace are used to construct a time-dependent joint probability mass function. The joint probability mass function is the representation of the Pzrecewise Independent Stochastic Process model for the trace. Two methods that assist to compact the model, while retaining accuracy, are also discussed.
  • Time Dependent Rate-Based Modeling for Trace Data
    Mathur, Anup; Abrams, Marc (Department of Computer Science, Virginia Polytechnic Institute & State University, 1995-10-01)
    This paper introduces a novel technique to construct an empirical model fitting time-varying (transient) trace data. The trace is a categorical or numerical time-series. The modeling technique described in this paper first builds a Mass Evolution Graph (MEG) from the trace data. Linear regression is then used to construct a time-dependent probability mass function (pmf) for each state in the trace data.
  • Thumbnail Image
    Toward a Machine Assisted Software Performance Diagnosis Methodology
    Mathur, Anup; Abrams, Marc (Department of Computer Science, Virginia Polytechnic Institute & State University, 1993-04-01)
    This paper discusses a methodology for diagnosing performance problems for parallel and distributed programs. The methodology is based on the formation and testing of hypotheses about the cause of performance bottlenecks. The process is illustrated with a case study of an actual problem arising in a parallel discrete event simulation program in which granularity is a primary bottleneck and barrier implementation is a secondary bottleneck. The paper also describes the evolution of Chitra, a software performance measurement and analysis tool whose objective is to automate certain steps in software performance diagnosis.
  • Visualizing and Modeling Categorical Time Series Data
    Ribler, Randy; Mathur, Anup; Abrams, Marc (Department of Computer Science, Virginia Polytechnic Institute & State University, 1995-06-01)
    Categorical time series data can not be effectively visualized and modeled using methods developed for ordinary data. The arbitrary mapping of categorical data to ordinal values can have a number of undesirable consequences. New techniques for visualizing and modeling categorical time series data are described, and examples are presented using computer and communications network traces.