The Scalability of Flwo Control Algorithms for ABR service in ATM Networks

摘要

While a number of flow control algorithms have been proposed and many simulation studies have been dome during the evolution of ATM Forum's rate based flow cntrol framework for ABR (Available Bit Rate) service in ATM networks, few of the proposed algorithms have been examined on such a large scale as today's Internet. In this paper, we study the scalability of distributed maxmin fair rate allocation (DMFRA), the flow control algorithms for ABR service with respect to convergence time. While our results are given only for the CCERI (Congestion Control with explicit Rate Allocation) algorithm, similar results can be obtained for some other DMFRA algorithms as well. We first propose a new computational model of maximin fair rate allocation by applying the theoretical results presented in [10]. Guided by this model, we obtain a new and much tighter upper bound on the convergence time of CCERI. With these results, we then try to address the scalability of the algorithm with respect to convergence time. We propose a network model to capture the characteristics of network growth and derive an upper pound on the convergence time of it in this scenario. We show that CCERI scales well as networks grow with number of links.