A Proposal of a Distributed Flow Control with Backward Propagation

摘要

Control of a large-scale network using a centralized approach is essentially difficult due to its large end-to-end delay, high heterogeneity of a large number of network components, low availability and/or reliability caused by network component failures. In this paper, using an autonomous and distributed approach, we aim at realizing a control mechanism, which is scalable in terms of the network size, for joint optimization of the multi-path routing and bandwidth allocation (MRBA). In this paper, we propose a distributed and scalable flow control mechanism called DFC-BP (Distributed Row Control with Backward Propagation), which simultaneously solves multi-path routing and bandwidth allocation. DFC-BP is an autonomous and decentralized hop-by-hop flow control mechanism which can minimize the total network cost utilizing the backward propagation from downstream nodes to upstream nodes. We also investigate the effectiveness of DFC-BP in terms of efficiency, transient performance, adaptability, and parameter sensitivity through simulation experiments. Our findings include that the total network cost realized by DFC-BP is comparable to that by a centralized heuristic algorithm, and that DFC-BP quickly adapts to the occurrence of multiple link failures.