Why Rely on Blind AIMDs?

摘要

We are motivated by the fact that fixed Increase rates and Decrease ratios for AIMD cannot adjust TCP's performance to the Internet's diverse networking conditions. Indeed, we find that fixed values for the increase/decrease factors of AIMD restrain flexibility, which is a fundamental property of transport protocols in order to guarantee utilization and fairness in Modern and Future internetworks. We propose a new paradigm for hybrid AIMD designs that has the potential to adjust TCP's behavior according to network conditions.rnThe proposed Multi-Rate AIMD (MR-AIMD) increases additively the Additive Increase factor of AIMD in case of positive feedback and decreases multiplicatively (the AI factor) in case of negative feedback and Explicit Congestion Notifications. In other words, MR-AIMD takes into account ECN signals in order to quantify the level of network contention and adjusts its response accordingly.rnWe show that MR-AIMD reduces retransmission effort significantly, when contention is high, becomes aggressive when contention decreases and tolerates against random, transient errors due to fading channels.