A randomized algorithm for vote assignment is described. Given a collection of sites where copies of a file are replicated, and the individual site reliabilities, the algorithm assigns votes to sites so as to maximize overall availability. It is based on the concept of simulated annealing which has been successfully applied for various complex problems (for instance, the traveling salesman problem) for which no efficient optimal algorithm is known. The authors tested their algorithm for 5-, 6-, 7-, 8-, and 9-site examples and compared the vote assignments produced by it with those produced by an optimal algorithm based on solving an integer programming model. The results produced by the randomized algorithm were remarkably close to those produced by the optimal one.