An analytical model of fast retransmission and recovery in TCP-SACK

Fast retransmission and recovery are the two most important mechanisms employed by TCP to timely recover lost packets and efficiently improve performance. This paper presents a mathematical model to systematically analyze the characteristics of fast retransmission and recovery in TCP-SACK, while multiple packet losses in a congestion window are possible. One of the significant observations revealed by this paper is that whether or not the lost packets can be fast retransmitted at the (y+1)(y+1)-th RTT (Round-Trip Time) round simply depends on Nr[y]Nr[y], representing the number of packets following the first lost packet and those packets have successfully arrived at the receiver at the yy-th RTT round. For clarification, the analytical model consists of two cases. The first case assumes that Nr[y]Nr[y] is greater than or equal to a threshold kk (the number of duplicate acknowledgments), while the second case investigates the behavior when Nr[y]Nr[y] is smaller than the threshold kk. In the proposed theorems, we show that there is an upper bound in the number of packet losses, above which TCP-SACK may not be able to recover the lost packets successfully. Additionally, we develop a model to study the throughput degradation resulting from multiple packet losses in TCP windows. The analytical results of throughput degradation are verified through OPNET simulation.