| Summary: | 碩士 === 國立臺灣大學 === 資訊管理研究所 === 90 === TCP (Transmission Control Protocol) is by far the most widely used transport layer protocol. It provides a connection-oriented reliable service to many of Internet applications. However, the dynamic TCP behavior results in some drawbacks such as reduced throughput or unfairness between TCP connections. Moreover, for a corporate that connects to the Internet through an access link with limited bandwidth, a traffic management mechanism is needed to provide different quality of service among TCP connections because some connections are mission critical while others are not.
Previous research shows that TCP’s throughput is inversely proportional to round trip time and square root of packet loss rate. In this paper, we described an end-to-end TCP throughput control mechanism, LTRCS. The idea is to controls TCP throughput through introducing extra packet loss rate or round trip time. The control scheme doesn’t manipulate the content (TCP header) of the packets, thus it is applicable even if the packets are authenticated or encrypted. Furthermore, LTRCS could control other “TCP-friendly” transport layer protocols which uses packet loss rate to estimate the available bandwidth of the network.
At last, LTRCS with Per Class Control which controls aggregate throughput of a class of TCP connections is also proposed to reduce the overhead of rate enforcement. Simulation results show that LTRCS gives good throughput control performance for both homogeneous and heterogeneous TCP connections.
|
|---|
