A server placement algorithm conscious of communication delays and relocation costs

Abstract

The server placement algorithm is to locate the given number of cache servers at "proper" coordinates in the network. A typical objective to determine good locations may be simply to find the set of client clusters in which the Euclidean center of each cluster is the location of the cache server. We claim, however, that the objective should also consider 1) the network communication delays and 2) the cost of relocating cache servers, if any. We exploit both hierarchical and partitioning approaches, and present our server placement algorithm. We evaluated the performance of the algorithm, and its result is promising.