隨著網路訊務流量的快速成長，如何妥善的運用網路資源是一個成功的擁塞控制機制要面對的根本問題。在終端設備上執行的TCP是網路上最廣為使用者使用的傳輸層協定，它有很多不同的版本被設計出來改進使用的效能，例如TCP Reno、TCP Vegas 等。由於TCP所棲身的終端設備並未具有網路內部狀態的資訊，大部份的TCP 擁塞控制機制僅能依賴封包遺失觸發擁塞控制機制，本研究提出TCP Muzha協定，藉由路由器協助，提供網路內部資訊給傳送端，在未發生擁塞前不需依賴封包遺失便可進行適度的傳輸速度控制，以減少因為封包遺失所造成劇烈的傳輸速度下降，並可更快速達到最佳傳輸速度。本研究的設計理念是設法尋找傳送路徑中的瓶頸，進而計算出瓶頸提供的可用頻寬，藉由瓶頸所提供的資訊動態的進行流量控制以充份利用頻寬並避免產生擁塞，增進整體的效能。本研究之重點在於路由器應提供何種資訊及如何運用所獲得的資訊進行動態速率調整。我們提出模糊化的多層級速率調整方法，藉著動態所獲得的細膩資訊做擁塞避免。最後於NS2平台實驗模擬，評估我們所提出的方法，實驗結果中顯示本方法能有效避免擁塞的產生，降低封包遺失，提升整體效能，和TCP Reno共存的環境下不因為Reno侵略性的傳輸方式而降低過多的效能並保有較低的封包遺失率。 / With the tremendous growth of Internet traffic, to utilize network resources efficiently is essential to a successful congestion control protocol. Transmission Control Protocol (TCP) is a widely used end-to-end transport protocol across the Internet. It has several enhencing versions (i.e. TCP Reno, TCP Vegas…) which intend to improve the drawbacks of the initial version of TCP. Most congestion control techniques use trial-and-error-based flow control to handle network congestion. In this paper, we propose a new method (TCP Muzha) that requires routers to feedback their status to the sender. Based on this information, the sender is able to adjust the sending data rate dynamically. Our approach can prevent data rate from decreasing dramatically due to packet loss. It can also help to increase the data rate quickly to where it supposes to be. Our design philosophy is to find out the bottleneck of the path, and its available bandwidth. Our goal is to increase network performance and avoid congestion by using the information obtained from the bottleneck. The design challenges are to determine which information is essential and how to use this information to dynamically adjust the data rate. We also propose the multi-level data rate adjustment method. Congestion can be avoided by dynamically adjusting data rate using this information. Finally, we use NS2 simulator to evaluate the performance of our approaches. From the experiment results, it shows our method can avoid congestion before it actually happen, decrease packet-loss rate and increase the network utilization. In the fairness experiment, our method will only suffer a minor throughputs decreasing when TCP Reno is coexisting.

隨著網路訊務流量的快速成長和無線網路技術日漸成熟，如何妥善的運用有限的網路資源是一個成功擁塞控制機制要面對的根本問題。TCP為現行網路上最廣為使用的傳輸層協定，並且有許多的不同版本被提出來改進其效能上的問題，例如TCP NewReno，TCP SACK 及TCP Vegas等。然而由於TCP傳送端並未具有網路內部狀態的資訊，如可用頻寬等，大部份的TCP擁塞控制機制僅能依賴封包遺失做為觸發擁塞控制的指標。許多研究指出在無線的環境下TCP無法有效使用有限的資源並且分辨封包遺失的原因，因而造成整體的效能不佳。本篇研究提出一個藉由路由器輔助的TCP擁塞控制協定－TCP Muzha，仰賴路由器提供調速資訊，以幫助傳送端能不依靠封包遺失進行傳輸速度控制，並可更快速的達到最佳的傳輸速度。本研究同時提出模糊化的多層級速率調整方法，藉由動態所獲得的細膩資訊做擁塞避免及因應無線環境下因路由改變或傳輸介質不穩所產生的不必要傳輸速度減低。最後我們在NS2模擬器上對所提出的協定做效能評估，實驗結果顯示本協定除了能有效的避免擁塞外，並能減少不必要的降速及重傳封包的次數。 / Communication networks have evolved tremendously in the past decades. TCP is the most dominant and deployed end-to-end transport protocol across Internet today and will continue to be in the foresee future. It has numerous enhancing versions for wired network such as TCP Reno, TCP NewReno and TCP Vegas to improve the drawbacks of initial version of TCP. As IEEE 802.11 wireless network technology gains popularity, TCP is very likely to be popular for existing applications so far. However due to unawareness of network conditions, regular TCP is not able to fully control the limited resources and distinguish packet loss from congestion loss and random loss. Based on such implicit assumption, many studies have shown this would results in serious performance degradation in wireless environment. In this paper, we proposed a new TCP congestion control mechanism by router-assisted approach which is inspired by the concept of each wireless node playing the roles of terminal and router simultaneously. Based on the information feedback from routers, sender is able to adjust the sending speed dynamically in order to avoid overshooting problem. We also proposed a multilevel date rate adjustment method to control the date rate more precisely. Finally we evaluate the performance of our approach by NS2 simulator. Our proposed protocol has 5~10% higher throughput than TCP NewReno and much less number of retransmission. The fairness requirement is also achieved while our proposed protocol coexists with other major TCP variants.