Bridging the gap between simulation and experimental evaluation in computer networks

Simulation and experimentation are two complementary techniques for performance evaluation, each one of them having opposite characteristics and advantages. While one allows total control and abstraction in the experiment, the other provides greater detail and realism. Though ideally it would be desirable to perform both, it is hard to direct the required efforts to develop a performance evaluation twice, once over a simulator and then again on a real network. In this paper, a new approach is explored through a tool called Simmcast testbed, which allows one to execute, from a single codebase, experiments both in simulation and experimentation mode. A didactic example is discussed in detail, and the correlation of the simulated and experimental results is presented.