In this paper the problem of maximal increase of system reliability is formulated as a resource allocation problem under a budget constraint. Dynamic programing is used for the optimal solution. Time to system failure is dictated by a Markov process. The system is composed of several subsystems. Each subsystem has several possible configurations that exhibit different levels of fault tolerance and incur different incremental costs at different times. Configuration dependence among subsystems is allowed. An example resembling a fault tolerant industrial process is presented, for which the proposed algorithm is used to obtain a set of maximally reliable solutions corresponding to a set of budget constraints.