1st International ICST Workshop on Network Simulation Tools

Research Article

Speed and Accuracy of Network Simulation in the SimGrid Framework

Download779 downloads
  • @INPROCEEDINGS{10.4108/nstools.2007.2010,
        author={Kayo Fujiwara and Henri Casanova},
        title={Speed and Accuracy of Network Simulation in the SimGrid Framework},
        proceedings={1st International ICST Workshop on Network Simulation Tools},
        proceedings_a={NSTOOLS},
        year={2010},
        month={5},
        keywords={},
        doi={10.4108/nstools.2007.2010}
    }
    
  • Kayo Fujiwara
    Henri Casanova
    Year: 2010
    Speed and Accuracy of Network Simulation in the SimGrid Framework
    NSTOOLS
    ICST
    DOI: 10.4108/nstools.2007.2010
Kayo Fujiwara1,*, Henri Casanova1,*
  • 1: Information and Computer Sciences Dept. University of Hawai‘i at Manoa 1680 East-West Rd, POST 317 Honolulu, HI 96822, U.S.A.
*Contact email: kayof@hawaii.edu, henric@hawaii.edu

Abstract

The experimental study of distributed systems and algorithms for large-scale internet platforms typically requires simulation of compute and network resources. This paper focuses on network simulation issues. While many packet-level network simulators are available and enable high-accuracy simulation, they can lead to prohibitively long simulation times. Consequently, a number of simulation frameworks have been developed that simulate networks at higher levels, thus enabling fast simulation but losing accuracy. One such framework, SimGrid, uses a flow-level approach that approximates the behavior of TCP networks, including TCP's bandwidth sharing properties. In this paper we perform a quantitative evaluation of SimGrid and compare it to popular packet-level simulators. We identify the regimes in which SimGrid's accuracy is comparable to that of these packet-level simulators, and the regimes in which SimGrid's accuracy may not be acceptable. We then describe an integration of the GTNetS packet-level simulator and SimGrid, which allows SimGrid users to easily opt either for fast but potentially inaccurate flow-level simulation or for accurate but potentially prohibitively slow packet-level simulation.