Can API-RCP achieve max-min fair bandwidth allocation in a multi-bottleneck network?

摘要

Theoretical analysis and experiments have shown that TCP can become oscillatory and prone to instability in high bandwidth-delay-product networks. XCP was proposed to overcome these shortcomings by advertising an explicit window adjustment from the routers to the sources. However, XCP may under-utilize the bandwidth in a multi-bottleneck network and cause some flows to receive an arbitrarily small fraction of their max-min fair rates. Using solid control theoretical analysis and design, API-RCP has solved this potential problem of XCP successfully. Why API-RCP can achieve max-min fair bandwidth allocation and full link utilization in steady state in multi-bottleneck networks? To address this question, we made a simple theoretical analysis and then verify it by OPNET^reg simulations. We also propose a controller design scheme to optimize the system performance with minimum control errors. We demonstrate the improved performance through comparison between API-RCP and TCP/RED in a multi-bottleneck network.