TY - CHAP
T1 - Analytic models of loss recovery of TCP Reno with packet losses
AU - Kim, Beomjoon
AU - Lee, Jaiyong
N1 - Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 2003
Y1 - 2003
N2 - In this paper, we investigate the loss recovery behavior of TCP Reno over wireless links in the presence of non-congestion packet losses. We consider both random and correlated packet loss, and derive the conditions that packet loss can be recovered without retransmission timeout (RTO) by accurate modeling of loss recovery behavior of TCP Reno. Through probabilistic work with the conditions derived, we compute the fast retransmit probability for packet loss probability. According to our results, only 25% of two packet losses in a window can be recovered by two fast retransmits. In a particular case, three lost packets can be recovered by fast retransmits, but its probability is extremly low. Since more than four packet losses in a window can be recovered by fast retransmits in no cases, RTO always occurs. The continuity of correlated packet losses as well as packet loss rate can affect the fast retransmit probability. Even if overall packet loss probability is very low, successive packet losses can degrade the fast retransmit probability. We explain some of these observations in terms of the variation of the average window size with packet loss probability.
AB - In this paper, we investigate the loss recovery behavior of TCP Reno over wireless links in the presence of non-congestion packet losses. We consider both random and correlated packet loss, and derive the conditions that packet loss can be recovered without retransmission timeout (RTO) by accurate modeling of loss recovery behavior of TCP Reno. Through probabilistic work with the conditions derived, we compute the fast retransmit probability for packet loss probability. According to our results, only 25% of two packet losses in a window can be recovered by two fast retransmits. In a particular case, three lost packets can be recovered by fast retransmits, but its probability is extremly low. Since more than four packet losses in a window can be recovered by fast retransmits in no cases, RTO always occurs. The continuity of correlated packet losses as well as packet loss rate can affect the fast retransmit probability. Even if overall packet loss probability is very low, successive packet losses can degrade the fast retransmit probability. We explain some of these observations in terms of the variation of the average window size with packet loss probability.
UR - http://www.scopus.com/inward/record.url?scp=21144432930&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=21144432930&partnerID=8YFLogxK
U2 - 10.1007/978-3-540-45235-5_92
DO - 10.1007/978-3-540-45235-5_92
M3 - Chapter
AN - SCOPUS:21144432930
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 938
EP - 947
BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
A2 - Kahng, Hyun-Kook
PB - Springer Verlag
ER -