τ-AIMD for asynchronous receiver feedback

In this work we first describe an asynchronous-feedback model which corresponds to the diverse round-trip times (RTTs) of competing flows within the same communication channel. We show that even when the feedback of the receivers/network is asynchronous, the duration of a congestion epoch represents "common knowledge" for the competing flows. Based on this property, we propose a diverse linear increase scheme in proportion to the RTT of each flow and we adjust periodically the windows of the competing flows accordingly, in order to enhance fairness of asynchronous-feedback environments. We name this scheme τ-AIMD and show that fairness of flows in asynchronous systems is improved since the increase-rate diversity of long and short RTT flows is largely canceled. We use the max-min notion of fairness to show that τ-AIMD can constitute the foundation for other window-based, end-to-end algorithms that target a good balance between efficiency and fairness.