Computer and Information Science and Engineering 

We consider the problem of analyzing the risk in a very large database of probabilistic events. We compute the risk as a function of the number of events that take place. We develop a novel algorithm that can accurately perform risk analysis for over one hundred million data points in only a few min- utes on commodity shared memory multiprocessors. Our algorithm precisely computes the full distribution of count over the occurrence of large number of uncertain events. To achieve this, it uses a novel approach that exploits the fast fourier transform for polynomial multiplication. We use our algorithm to compute the risk for the damage caused by hurricane for a variety of risk functions on over one hundred million homes and demonstrate the scalability and useful- ness of our approach. Comparisons with competing approx- imate methods that can scale to the such large probabilistic datasets suggest that our approximate methods can yield significant errors in computing the risk while providing lit- tle or no advantage in running time over our exact method.