Corruption-Based TCP Rate Adjustment in Wireless Networks

Xu Chang · Chinese Journal of Computers · 2002

Comparing to wire networks, there are many different characteristics in wireless environments, such as higher error rate, longer delay, lower bandwidth, frequent mobility and so on, which makes TCP congestion control mechanisms not directly suitable for wireless networks, and many improved TCP congestion control mechanisms have been presented. Generally, the influences to TCP sender's packet sending rate by link corruption are not considered in these improved schemes. However, unnecessary packet lost by corruption can be greatly avoided through the decrease of packet sending rate, which may lead to higher reliability, unnecessary energy consumption of mobile hosts, and less system overheads. For a given TCP connection, it is reasonable to assume that the possibility of packet lost by corruption can be obtained approximately from p e=m/n , where n is the number of total packets and m is the sum of packets lost by corruption during the period of time T . In this way, the possibility for n packets to be sent successfully to receiver is p=(1-p e) n , and TCP sender's maximum average packet sending rate is rate max =log a log( 1-p e ), where a is the lower limit of p . Based on the main idea, we obtain three TCP corruption based rate adjustment mechanisms, namely, two mechanisms driven by the event of packet loss in which TCP sender's packet sending rate is adjusted as soon as packet lost by corruption is detected, and one mechanism driven by TCP sender's sending period in which the rate is only adjusted while sending period expires. In terms of analysis, the mechanisms presented in this paper have the following characteristics: (1) TCP sender's packet sending rate is adjusted self adaptively by link corruption. (2) Unnecessary packets lost by corruption can be greatly avoided with light influences to throughput and delay through the choic of parameter a from (1-p e) rate 0×T/3 to (1-p e) rate 0×T/6 at the time of heavy packet loss by corruption. (3) The mechanisms can also be simply implemented with less overheads.

Read the paper · More papers on PaperTik