Cache Policies for Linear Utility Maximization