Efficient simulation experiments for comparing communication network routing algorithms

Abstract

In this paper, we propose efficient statistical designs of simulation experiments and statistical methods for systematically comparing communica tion network routing algorithms via simulation. The design of experiments technique establishes the minimum number of observations or simulation runs without losing much information needed to gain insight into the real world system. Statistical models use this efficient set of observations in providing (1) insight into the intrinsic behavior of the algorithm (i.e., sensitivity of the algorithm to changes in factors), (2) a statistical basis for com paring alternate algorithms that perform common functions, and (3) predictor equations that can be used to predict algorithm performance for the user- specified values of influencing factors. The empha sis in the paper is on the use of statistical testing and comparison methods in a simulation setting and not on the nature of algorithms themselves. These methods should pave the way for perform ance analysis of other algorithms as well.

The testing environment includes a parameter- driven simulator which has the capability to modularly include alternate routing algorithms one at a time. The model generates a threat scenario which transforms into the generation of appropriate messages. The impact analysis of five influencing factors on three measures of perform ance is proposed.