2nd International ICST Workshop on OMNeT++

Research Article

Parallelizing OMNeT++ Simulations using Xgrid

  • @INPROCEEDINGS{10.4108/ICST.SIMUTOOLS2009.5554,
        author={Robin Seggelmann and Irene R\'{y}ngeler and Michael T\'{y}xen and Erwin P. Rathgeb},
        title={Parallelizing OMNeT++ Simulations using Xgrid},
        proceedings={2nd International ICST Workshop on OMNeT++},
        publisher={ACM},
        proceedings_a={OMNET++},
        year={2010},
        month={5},
        keywords={OMNeT++ Xgrid Cluster},
        doi={10.4108/ICST.SIMUTOOLS2009.5554}
    }
    
  • Robin Seggelmann
    Irene Rüngeler
    Michael Tüxen
    Erwin P. Rathgeb
    Year: 2010
    Parallelizing OMNeT++ Simulations using Xgrid
    OMNET++
    ICST
    DOI: 10.4108/ICST.SIMUTOOLS2009.5554
Robin Seggelmann1,*, Irene Rüngeler1,*, Michael Tüxen1,*, Erwin P. Rathgeb2,*
  • 1: Münster University of Applied Sciences, Fachbereich Elektrotechnik und Informatik, Stegerwaldstrasse 39 D-48565 Steinfurt, Germany.
  • 2: University of Duisburg-Essen, Institute for Experimental Mathematics, Ellernstrasse 29 D-45326 Essen, Germany.
*Contact email: seggelmann@fh-muenster.de, i.ruengeler@fh-muenster.de, tuexen@fh-muenster.de, erwin.rathgeb@iem.uni-due.de

Abstract

Working with simulations, testing and validating theories often requires a large number of simulation runs. The discrete event simulation environment OMNeT++ already provides functionality for distributed computing, yet the simulated model and modules to be parallelized have to be declared manually. Apple Mac OS X comes with Xgrid support, which allows easily setting up an ad hoc grid for parallel computing. In this paper we will describe the features we had to add to OMNeT++ to be able to use Xgrid to parallelize even several thousand runs. We will point out how to use Xgrid to distribute runs of a simulation not only to multiple CPU cores but also to multiple machines without modifying the simulation itself. Our analysis will reveal that Xgrid allows to reduce the computing time almost proportional to the added parallel computing power.