• DocumentCode
    1354395
  • Title

    A Performance Modeling Scheme for Multistage Switch Networks With Phase-Type and Bursty Traffic

  • Author

    Yu, Ming ; Zhou, MengChu

  • Author_Institution
    Dept. of Electr. & Comput. Eng., Florida State Univ., Tallahassee, FL, USA
  • Volume
    18
  • Issue
    4
  • fYear
    2010
  • Firstpage
    1091
  • Lastpage
    1104
  • Abstract
    Existing analytical methods to model multistage switch networks cannot be applied to the performance modeling of switch networks with phase-type and bursty traffic because of the problem of state-space explosion and unrealistic assumptions, e.g., uniform traffic and independent destination (UTID). This paper presents an approximate scheme to model and analyze such networks. First, a traffic aggregation technique is proposed to deal with phase-type and bursty traffic, including splitting and merging. For the aggregation of two bursty traffic, a closed-form solution is obtained for buffer state probabilities. For the aggregation of more bursty traffic, a recursive algorithm is derived in terms of the buffer size and number of inputs of a switch. Second, a switch decomposition technique is developed, by which the crossbar of a switch is decomposed from its preceding and succeeding buffers. In this way, a switch network of N inputs and outputs is converted to N tandem queues, for which the performance can be easily evaluated. Our extensive numerical and simulation examples have shown that the proposed scheme achieves satisfied accuracy and computational efficiency.
  • Keywords
    optical switches; packet switching; probability; telecommunication traffic; buffer size; buffer state probability; bursty traffic; merging; multistage switch network; packet switch; phase-type; recursive algorithm; splitting; state-space explosion; switch crossbar; switch decomposition; traffic aggregation; uniform traffic and independent destination; Finite buffer; performance modeling; switch network; traffic modeling;
  • fLanguage
    English
  • Journal_Title
    Networking, IEEE/ACM Transactions on
  • Publisher
    ieee
  • ISSN
    1063-6692
  • Type

    jour

  • DOI
    10.1109/TNET.2009.2036437
  • Filename
    5352328