A Generic Scheme for Estimating the Size of a Group While Avoiding Feedback Implosion

We present a probabilistic polling scheme for estimating the size of a group of nodes affected by the same event. We analyze the bias of the proposed scheme and show how it can be completely eliminated. Our scheme differs from previous work in several important ways. First, it is generic in the sense that it is not dependent on the physical properties of the underlying network. Second, it uses a “one-shot” estimation technique that does not depend on the results of previous rounds. Finally, the estimating nodes control the number of feedback messages, thus allowing a good balance between the overhead imposed by the scheme and its precision. We compared our scheme to previous works and saw a notable performance improvement: our scheme reduced the number of response messages by 80-90% while obtaining the same precision.