Simulating Large Parallel Machines for Performance Prediction

The masters thesis describes a simulator for large parallel machines like BlueGene/L that provides the ability to make performance predictions, based on actual execution of real applications. It is useful to evaluate the performance of many applications on such machines even before they are built. In this thesis, we present an online parallel simulator which is based on a parallel discrete event simulation scheme and also demonstrate sequential and parallel post-mortem simulation schemes. The techniques for optimizing online parallel simulations of machines with large number of processors on the ones with fewer number of processors are also explored. Finally, a comparison of online and post-mortem approaches is also presented.

Gunavardhan Kakulapati, "Simulating Large Parallel Machines for Performance 
Prediction", University of Illinois at Urbana-Champaign, 2003.

@mastersthesis{GunaMSThesis03,
        author = "Gunavardhan Kakulapati",
        title = "Simulating Large Parallel Machines for Performance Prediction",
        year = "2003",
        school = {University of Illinois at Urbana-Champaign},
}