長期演進行動網路上透由傳輸控制協定之擁塞控制改進的延遲降低

Abstract

延遲是一個關鍵的網路效能指標，因其對於使用者經驗有顯著的影響。為了降低延遲以求更好的使用者經驗，我們著眼在現今行動網路中最必不可少且廣泛使用的通訊協定之一─傳輸控制協定。在傳輸控制協定中，主要影響傳輸速率的是擁塞控制的機制。在過去，擁塞控制被發現在無線的環境中，會不必要地降低擁塞窗口，而造成額外的延遲。因此，九零年代晚期以降，許多改善機制被提出。然而，相較過去，現今的數據鏈路層較為可靠，而將大部分的傳輸損耗處理得不讓上層察覺，使得這個在傳輸控制協定中的問題能被預防。在本篇論文中，我們指出在某些狀況下，即使在可靠的數據鏈路層上，這個問題還是有可能發生。因此，我們提出一個透由長期演進技術協助來防止在這些狀況下非必要降低擁塞窗口的機制。並且以模擬來顯示其相對於既有傳輸控制協定中擁塞控制的改善。

Latency is a key network performance measure, since it has tremendous effect on the user experience. In order to reduce the latency for better user experience, we focused on the Transmission Control Protocol (TCP), one of the most essential and widely used protocols in the mobile network today. In the TCP, the mechanism of congestion control mainly affects the sending rate. Congestion control was found to unnecessarily decrease the congestion window (cwnd) in the wireless environment and cause additional latency. Hence, many solutions were proposed in the late 1990s. However, the modern data link layers are more reliable that make most transmission losses invisible to the upper layers and prevent this problem in the TCP. In this thesis, we pointed out that this problem may still exist even over a reliable data link layer in some cases. Hence, we proposed a mechanism to prevent unnecessary cwnd decrement in such cases with the help of the LTE, and performed simulation to show the improvement over the original TCP congestion control.