Large Distributed databases are often subject to weaker query optimization due to system complexity. Query execution requires data transfers between the distant processing sites of the system. In this paper we propose a solution for minimizing raw data transfers between distant nodes by online re-arranging and replicating data within the constraints of the original database architecture. Data transfer is the principal factor when transferring intermediate results in the process of query evaluation. The proposed method gathers online incremental knowledge about data access patterns and database statistics to solve the following problem: online re-allocation of the fragments in order to constantly optimize the query response time. We model our solution as a mathematical linear programming problem and show in the final section the experimental results we obtain by comparing the improvements obtained between various database configurations, before and after optimization.