Deriving Fast Recovery Probability as a Performance Measure of TCP SACK in Wireless Links

In this paper, we would like to study the analytical model of Transmission Control Protocol – Selective Acknowledgement (TCP SACK) by deriving its fast recovery probability in wireless links. Recently, there exist several variants of TCP, for instances Tahoe, Reno, New Reno, Vegas, Westwood and SACK. TCP Reno is the most popular TCP implementation in the Internet today. Reno’s fast recovery algorithm is optimized for the case when a single segment is lost from a window. However, Reno can suffer from performance problem when multiple segments are lost in a window and generally recovers the segment losses by waiting retransmission time out. Meanwhile, TCP Selective Acknowledgment (SACK) aims at providing efficient retransmission in the presence of multiple segment losses. SACK is proposed to enhance the performance of Reno in the case of multiple losses in a window. Most of studies discussed the performance of TCP SACK under correlated losses since the researchers assumed the correlated losses scenario to be a dominant flow model in the Internet due to widely implementation of drop-tail queue routers. On the other hand, because of interleaving process, the error model in the wireless links can be modeled into random losses. We stochastically developed analytical model of SACK under random losses by deriving a fast recovery probability as a significant parameter to know how efficient SACK recovers the losses without waiting a retransmission timer expires.