Multicast session membership size estimation

Abstract

We derive estimators and bounds that drive probabilistic polling algorithms for the estimation of the session size, n, of any potentially large scale multicast session. We base our analysis upon a mapping of polling mechanisms to the problem of estimating the parameter n of the binomial (n,p) distribution. From the binomial model, we derive an interval estimator for n, and we characterize the tradeoff between the estimator's quality and its overhead in a manner readily matched to application requirements. We derive other estimators and bounds that enable applications to treat as a tunable parameter the confidence that they will not exceed their overhead limits. We also suggest revised estimators and other improvements for the mechanisms proposed by Bolot, Turletti and Wakeman (1994), and Nonnenmacher and Biersack (see Proceedings of IEEE INFOCOM '98, Los Alamitos, California, IEEE Computer Society Press, 1998).