This paper addresses the problem of software architecture according to reliability forecasting. In a distributed computing environment a statistical model is used for determining the suitable algorithms related to performance requirements for each specific application. The proposed technique is based on a chain of automatic data collection, which has given us the possibility to adjust, during the execution, the strategy of fault management. We propose a monitoring scheme with flexibility. In this monitoring scheme we define a specific library for file accesses and conversation in order to keep track of files and communication usage. The pattern incorporates a self-contained architecture for distributed systems, which allows us to monitor the collection of the statistical data and to support the execution of the applications in a non-reliable operational environment. The main advantage of the proposed technique is that it gives us the possibility to design the software architecture according to the most recent requirements