Equivalence Conditions of Buffered and Bufferless Network Architectures

Abstract:
Buffers in network devices are expensive and undesired because of latency reasons. Nonetheless, existing packet networks are buffered for the purpose of acceptable link utilization and low loss rates at the same time. On the other hand bufferless network architectures are well known, ranging from old fashioned telephony up to optical burst switching. We show in theory the similarity between both buffered and bufferless network dimensioning and we illustrate in a number of examples to which extend this theoretical similarity results in comparable technical performance.