Accuracy analysis of long-run average performance metrics

B.D. Theelen, J.P.M. Voeten, Y. Pribadi

    Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review


    Before implementing a system with hardware and software components, performance modelling is often used for evaluating design alternatives. The formal modelling language Parallel Object-Oriented Specification Language (POOSL) has proven to be very useful for analysing performance metrics of design alternatives for real-life industrial systems. Based on its mathematically defined semantics, POOSL enables to analytically compute performance metrics by analysing the Markov chain that is implicitly defined by a POOSL model. Models of real-life industrial systems are however often too complex to be analysed exhaustively. Performance evaluation is therefore based on simulation, enabling the estimation of performance metrics. However, simulation results only have a proper meaning if their accuracy is known. In general, longer simulation yields more accurate results. To analyse the accuracy of simulation results and automatically terminate the simulation after accurate results have been obtained, confidence intervals can be used. Based on the properties of the Markov chains implicitly defined by POOSL models, this paper presents a technique of regenerative cycles for analysing the accuracy of long-run sample average estimation. Furthermore, an algebra of confidence intervals is defined to enable accuracy analysis of long-run time averages as well as long-run sample and time variances. Finally, library classes are introduced for POOSL, which ease accuracy analysis for the mentioned performance metric types.
    Original languageEnglish
    Title of host publicationProceedings 2nd PROGRESS Workshop on Embedded Systems (Utrecht, The Netherlands, October 18, 2001)
    EditorsF. Karelse
    PublisherSTW Technology Foundation
    ISBN (Print)90-73461-26-X
    Publication statusPublished - 2001


    Dive into the research topics of 'Accuracy analysis of long-run average performance metrics'. Together they form a unique fingerprint.

    Cite this