Throughput modeling of TCP with slow-start and fast recovery

Despite the rich literature on modeling TCP, we find two common deficiencies with the existing approaches. First, none of the work gives sufficient treatment to slow-start, although almost all of them show that retransmission timeout events are common. Second, the probability that retransmission timeout occurs has been underestimated, because retransmission timeout is coupled with fast recovery but fast recovery has not been properly modeled in the previous work. In this paper, new analytical models for predicting the steady state throughput of TCP flows are proposed. All major TCP mechanisms, including slow-start, congestion avoidance, fast retransmit, and fast recovery, are jointly considered under both bursty and independent loss models. We show that our proposed throughput models capture TCP performance more accurately.