from server i to server j). D models an underlying network topology. For our analysis we assume that the distances are symmetric and the triangle inequality holds on the distances (for all servers i, j, k: dij + djk ≥ dik). Each server has demand from clients that is represented by a demand matrix W (i.e., wij is the demand of server i for object j). When a server caches objects, the server incurs some placement cost that is represented by a matrix α (i.e., αij is a placement cost of server i for object j).
In this study, we assume that servers have no capacity limit. As we discuss in the next section, this fact means that the caching behavior with respect to each object can be examined separately. Consequently, we can talk about configurations of the system with respect to a given object: