Internet Cache Location and Design of Content Delivery Networks

Abstract

The paper studies the problem of where to locate caches in a network of a general topology with many servers. The problem is formulated using mixed integer programming (MILP). The goal is to develop new models for cache location that overcome the limitations of the basic model. A secondary goal is to evaluate the practical complexity of using MILP for cache location and to improve existing heuristics using the new model formulations.

The basic CLP model is studied as a multi-criteria problem to address the possibility that some servers or clients may be discriminated. The basic model is modified to allow searching for efficient solutions according to user preferences, for example for fair solutions. Using multi-criteria methods, the CLP can be extended to determine the optimal number of caches or to minimize average delay and bandwidth consumption. The paper studies modifications of the CLP that account for the effect of client assignment and cache size on cache hit rate.

Medium-sized topologies of the order of 100 nodes can be solved optimally using modern MILP solvers. The model modifications can be used to improve heuristics for larger networks.