一個增強型的頻寬時延乘積的擁塞控制方法

Abstract

暫存器膨脹是一種因為網路中存在一些過大的暫存器導致資料傳輸延遲會嚴重上升的問題。最近，GOOGLE為了解決暫存器膨脹，提出了一個全新的TCP協定叫做頻寬時延乘積的擁塞控制方法。雖然此方法可以解決此問題，但還有一些問題。當它與基於封包遺失的擁塞控制方法一起做競爭，它會被競爭者餓死。在我們的演算法中，我們可以判斷是否有競爭者存在，然後調整競爭的力度。從我們的模擬結果中，我們的傳輸速率在競爭的環境下比原本的機制還要好。在沒有競爭者的時候，也可以保持原先的行為維持最大傳輸速率並且降低延遲。

Bufferbloat is a phenomenon that will cause extremely high latency due to the oversized buffer deployed in the network. Recently, Google proposed a new TCP BBR in order to solve bufferbloat problem. Although the new TCP could solve this problem, but there is still some drawback. When BBR competes with loss-based congestion control, and it will be starved to death since it is too conservative relative to loss-based TCP. In our algorithm, we make a judgment whether there is any competitive traffic, and adjust the intensity to compete with other flows. From our simulation, compare to original BBR, our algorithm can have better throughput than original when there are other flows. When a single flow, we could keep BBR’s original behavior, then achieve maximum throughput and reduce latency.