Comparative Performance Analysis of Basic Window Size Adjustment Algorithms for Dynamic Windows Flow Controlled Internet Environment: Markov Renewal Reward Process Modeling

T. Ikegawa (Japan)


Dynamicwindow flow control, Window size adjustment algorithm, Comparative performance analysis, Renewal reward process, TCP


In dynamic-window flow control, a sender decreases the window size on detection of packet-loss and in creases the window size on acknowledgment of packet (re)transmission. The method’s ease of implementation and effectiveness in handling congestion in net works are well-known. This paper presents an analytical method for investigating the performance (mean effective throughput) of basic algorithms for increasing and decreasing the window size (referred to as window size adjustment algorithms), along with some results of the method’s application. To derive the mean effective throughput, we propose a renewal reward process that reflects the behavior of window size dynamics. By com paring the performances of window size adjustment algorithms operating with various error-recovery procedures, we show that a good algorithm is one where a binary-scheme for window size reduction is combined with a parabolic-scheme for size increase. This is the approach which has been implemented in the congestion avoidance phase of the popular Reno version of the TCP.

Important Links:

Go Back