在異質網路上用以改善傳輸控制協定之封包遺失分類演算法

Abstract

傳輸控制協定(TCP)是網路廣泛使用的傳輸協定。隨著無線網路的普及，傳輸控制協定開始被使用在無線網路上。然而傳輸控制協定所包含的擁擠控制演算法是以封包遺失當成網路擁擠的指標，在有線及無線網路上除了有因為擁擠造成的封包遺失還有因為無線網路訊號衰減或受到屏蔽造成的封包遺失。如果擁擠控制演算法把無線網路造成的封包遺失當成網路擁擠的指標，錯誤地減少傳送速度，將會造成不必要的效能衰減。 在本論文中，提出一個封包遺失分類演算法，以相對的傳送時間當成分類的依據將封包分為擁擠造成的遺失或是無線網路造成的遺失，使得擁擠控制演算法只對因為擁擠造成的封包遺失產生減低傳送速度的反應，因而可以避免上述因為錯把無線網路造成的遺失當成網路擁擠指標而發生的不必要的效能衰減。

Transmission Control protocol (TCP) is the most widely used transport layer protocol on Internet. As the popularity of wireless communication is on rise over the last few years, TCP is being extended to wireless network. However TCP is not suitable to be used on heterogeneous networks because its congestion control algorithm uses packet loss event as an indicator of network congestion. In a wired/wireless network, there are two classes of packet losses, wireless loss and congestion loss. Wireless loss is caused by common channel errors due to multipath fading, shadowing, and attenuation. Congestion loss is caused by network congestion. If the congestion control algorithm takes the wireless loss as an index of network congestion, it will mistakenly lead to dramatic performance. In this thesis, we propose a packet loss classification algorithm based on relative one-way trip time (ROTT) and use two trend detections to differentiate congestion loss from wireless loss in the ambiguous area of ROTT distribution. Then we show that our proposed algorithm can improve network performance comparing with other methods.