Designing an optimal overlay communication network for a set of processes on the Internet is a central problem of peer-to-peer (P2P) computing. Such a network defines membership and allows for members to disseminate information within the group. The network has to be robust and the available bandwidth has to be utilized in an optimal manner to allow for maximally efficient communication. This problem can be formulated as a dynamic optimization problem where classical combinatorial optimization techniques must face the further challenge of time-varying input data. ACO systems appear to be particularly fit for this class of problems, being able to construct an internal model of the instance to face and to exploit it for fast adaptation to modified contexts. This paper proposes to use elements resulting from mathematical techniques, in this case Lagrangean relaxation, in an ACO framework in order to achieve sound hot start states for fast response to varying network structures.