在多個TCP連線下解決暫存器膨脹之問題

Abstract

暫存器膨脹是一種因為過多的緩存資料而導致延遲提高、封包間延遲變大以及吞吐量下降的現象。然而現在研究這個議題的相關機構所使用的情境都不是針對多個傳輸控制協議。而且在多個傳輸控制協議連線下，會導致更嚴重的效能下降。因此，針對這個議題，我們提出了一個動態調整接收端窗口的機制。從我們的模擬結果中，相較於其他解決這個議題的方法來說，我們的演算法除了可以達到最大的吞吐量，同時也可以平均分配所有傳輸協議連線的頻寬。除此之外，我們也可以降低延遲來減輕暫存器膨脹的現象。

Bufferbloat is a phenomenon in which excessive buffering of packets cause high latency, larger jitter, as well as throughput degradation. However, the research groups who have paid attention to this issue do not focus on the scenarios under multiple TCP connections. In fact, the performance may seriously degrade under multiple TCP connections. To address this issue, we propose a mechanism which dynamically adjust the receive window. From our simulation results, our algorithm comparing to other solutions of bufferbloat issue can not only maximize throughput but also equally share the link capacity among all TCP connections at the same time. Besides, we also reduce the latency to mitigate the result bufferbloat causes.