The quality and reliability of software updates (SUs) are critical to a system vendor and its customers. As a result, it is important that SUs shipped to customers be successfully integrated into the field generic. A large amount of code must be shipped in an SU because customers want as many fixes and features as possible without compromising the reliability of their systems. However, as the size of an SU increases, so does its probability of field failure, thus making larger SUs riskier. The fundamental question is: How large should an SU be to keep the risk under control? This paper studies the tradeoff between the desire to ship large SUs and the failure risk carried with them. We formulate the problem as a nonlinear programming (NLP) problem, investigate it under various conditions, and derive sizing strategies for the SU. In particular, we derive a formula for the maximal SU size. We make a connection between software reliability and linear programming which, to the best of our knowledge, appears here for the first time. We also introduce some basic ideas related to the customer operational environment and explain the importance of the environment to software performance using an interesting analogy.