An Analytical Method for Predicting the Performance of Parallel Image Processing Operations

  • Authors:

  • Zoltan Johasz

  • Affiliations:

  • Department of Information Systems, University of Veszprem, PO Box 158, H-8201, Hungary E-mail: Zoltan Juhasz

  • Venue:
  • The Journal of Supercomputing
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents an analytical performance prediction model andmethodology that can be used to predict the execution time, speedup,scalability and similar performance metrics of a large set of imageprocessing operations running on a p-processor parallelsystem. The model which requires only a few parameters obtainable on aminimal system can help in the systematic design, evaluation andperformance tuning of parallel image processing systems. Using the modelone can reason about the performance of a parallel image processing systemprior to implementation. The method can also support programmers indetecting critical parts of an implementation and system designers inpredicting hardware performance and the effect of hardware parameterchanges on performance. The execution of parallel image processingoperations was studied and operations were arranged in three main problemclasses based on data locality and the communication patterns of thealgorithms. The core of the method is the derivation of the overheadfunction, as it is the overhead that determines the achievable speedup. Theoverheads were examined and modelled for each class. The use of the methodis illustrated by four class-representative image processing algorithms: image-scalar addition, convolution, histogram calculation and the Fast Fourier Transform. The developed performance model has been validated on a16-node parallel machine and it has been shown that the model is able topredict the parallel run-time and other performance metrics of parallelimage processing operations accurately.