In this paper we consider a large distributed system in which data is shared among several users. Specifically, we present a secure adaptive algorithm for data fragment allocation on multiple nodes of the system. The algorithm handles (replicated) data fragments, stored by nodes without the need of encryption, in such a way to ease information sharing. Data confidentiality is guaranteed in the presence of passive and active attacks, and fragments are dynamically reallocated/replicated in the system to converge, under assumptions of regularity of the read-write activity, to an allocation that provably guarantees highest performance in terms of network load.