Maya: a Multi-Paradigm Network Modeling Framework

  • Authors:

  • Junlan Zhou;Zhengrong Ji;Mineo Takai;Rajive Bagrodia

  • Affiliations:

  • University of California, Los Angeles, Parallel Computing Laboratory, 3809 Boelter Hall, Los Angeles, CA;University of California, Los Angeles, Parallel Computing Laboratory, 3809 Boelter Hall, Los Angeles, CA;University of California, Los Angeles, Parallel Computing Laboratory, 3809 Boelter Hall, Los Angeles, CA;University of California, Los Angeles, Parallel Computing Laboratory, 3809 Boelter Hall, Los Angeles, CA

  • Venue:
  • Proceedings of the seventeenth workshop on Parallel and distributed simulation
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents Maya, a multi-paradigm, scalableand extensible network modeling framework for emulatingdistributed applications. A novel three-tier architecture isproposed to integrate three disparate modeling paradigms,namely, discrete event models, analytical models and physicalnetwork interfaces into one unified framework of Maya.As the first effort to integrate all three paradigms intoone framework, this paper discusses the implementationsof Maya using Qualnet, fluid flow based TCP model andphysical network interface. It addresses the performanceissues involved in attaining the real time constraints imposedby distributed applications and demonstrates the effectivenessof using analytical models in Maya. Furthermore, it identifies the negative impact on real time performance through the computation intensive ordinary differential equation (ODE) solver in the fluid flow model. A new approach to interleaved executions of the fluid flow model is proposed to hide ODE solver turnaround time. As a result, the percentage of packets missing their deadlines has been reduced from more than 6% to less than 0.2%.