Simulating Large Networks { How Big is Big Enough?

Simulation has become the evaluation method of choice for many areas of computer networking research. However, most existing network simulation packages have severe limitations on the size and complexity of the network being modeled. Simulated networks of just a few thousand network elements and a few thousand data ows will quickly exhaust the computing resources in any reasonably sized computer workstation. Thus the researcher is faced with the dilemma of proving concepts designed to work eÆciently on networks of tens of millions of elements, using a simulation of only a few thousand elements. The grand challenge we discuss in this paper is that of using simulation to reach credible conclusions about Internet{scale network performance. We present data that demonstrates that simulation of Internet{scale networks is not presently feasible, nor is it likely to be feasible in the near future. We present a summary of current research in the eld of large scale network simulations. These recent advances, while not enabling Internet{scale simulations, do o er the tools with which one can begin to tackle the problem. We sketch one possible approach and describe the issues that need to be resolved in order to realize it.