Peer-to-peer (P2P) networks have been shown to be a natural and efficient paradigm for modeling most internet applications. However, data persistence remains still a major challenge, particularly in highly dynamic and unstable P2P networks. Within the framework of collaborative engineering, we propose a probabilistic approach based on peers collaboration to guarantee persistence of critical data in a system. Markov chains are used to model applications and realistically capture the behavior of practical systems. The model is first analytically investigated, and then data persistence is measured using an erasure coding redundancy scheme. The mathematical analysis allows us to determine the extent of data persistence in several important cases, and to anticipate the robustness of the large scale dynamic distributed applications.