Optimizing Capacity in Interconnection Networks with Finite Buffers

A great deal of recent research attention has been given to the throughput maximization of interconnection networks. These networks connect computer processing and routing resources via both physical and wireless links, where the resources are reconfigurable in a dynamic fashion. Jobs move through several processing stations and can be buffered between each stage. Most theoretical work assumes that the capacity for buffering at each stage is infinite, and general classes of scheduling policies have shown how to maximize the capacity under those conditions. Here we consider the case when such buffers are finite, and analyze the effect on network stability. We find that for very general arrival processes and an arbitrary fixed network topology, the stability region with finite buffers is a subset of that for the corresponding network with infinite buffers. As the capacity of buffers increases, the stability region eventually becomes equivalent to that for infinite buffer networks3.