基於Linux實現解決暫存器膨脹問題之I-D雙狀態演算法

Abstract

暫存器膨脹是一種因為網路中過大的暫存器導致資料傳輸延遲會嚴重上升的問題。先前已提出由兩個狀態的接收窗口調控機制來動態調整接收窗口以解決暫存器膨脹的問題，運用此演算法可以在降低延遲前先確保達到最高的傳輸效率，在此實做至Linux內核，並在實際網路環境測試及驗證。進而將兩個狀態轉換的概念移至發送端改成調控壅塞窗口，和其他發送端的解決方法比較，在吞吐量方面能達到較高的吞吐量，雖然在延遲方面無法一直維持在理想值，不過在網路有競爭情形的時候，此壅塞控制演算法可以得到較多分配。

Bufferbloat is a phenomenon that will cause latency to be extremely high because the larger buffer is deployed in the network. I-D two-state DRWA had been proposed with using dynamic receive window adjustment to solve bufferbloat issue. We can not only approach maximum bandwidth but also reduce the latency with the two-state mechanism. Here, we had implemented this algorithm into Linux kernel and used the real network as test environment, and the results matched to the previous simulations. Furthermore, we took this two-state concept into sender side. We built a new congestion control to mitigate bufferbloat. Compare to another sender-based solution, the performance of latency could not always be optimal. However, the throughput and fair-share with compete traffic are much better.