Optimal file sharing in distributed networks

The following file distribution problem is considered: Given a network of processors repre-sented by an undirected graph G = (V;E), and a file size k, an arbitrary file w of k bits is to be distributed among all nodes of G. To this end, each node is assigned a memory device such that, by accessing the memory of its own and of its adjacent nodes, the node can reconstruct the contents of w. The objective is to minimize the total size of memory in the network. This paper presents a file distribution scheme which realizes this objective for k ≫ log∆G, where ∆G stands for the maximum degree in G: For this range of k, the total memory size required by the suggested scheme approaches an integer programming lower bound on that size. The scheme is also constructive in the sense that, given G and k, the memory size at each node in G, as well as the mapping of any file w into the node memory devices, can be computed in time complexity which is polynomial in k and |V |. Furthermore, each node can reconstruct the contents of such a file w in O(k2) bit operations. Finally, it is shown that the requirement of k being much larger than log∆G is necessary in order to have total memory size close to the integer programming lower bound