A Feasible Baseline Architecture for Building and Evaluating Distributed Systems

Although distributed networks and systems are widely deployed, an effective overall benchmark architecture is still lacking. Recently we defined and studied the fat-stack network architecture and found that it is both efficient and scalable to distributed systems of various scales. In precedent work, we theoretically proved that the fat-stack is universal by measure of routing efficiency with a low 1.5-powered logarithmic overhead. Universality implies that the fat-stack of a given size is nearly the best routing network of that size. The fat-stack is also the minimal universal network in terms of number of links and nodes used. In this paper, we add more analytical proof and report our simulation results. Simulations show that the fat-stack outperforms a mesh-based distributed network of comparable hardware usage. We show how the fat-stack can be an effective benchmark architecture and how to scale the network from a VLSI graph layout to a large-scale distributed topology. Our work helps explain why some deployed networks function in the way they do in terms of routing. It also provides an exemplary network of proven efficiency and scalability for building new distributed systems.