Analytic models of loss recovery of TCP Reno with packet losses

Beomjoon Kim, Jai Yong Lee

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

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.

Original languageEnglish
Pages (from-to)938-947
Number of pages10
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2662
Publication statusPublished - 2003 Dec 1

Fingerprint

Packet Loss
Packet loss
Recovery
Loss Probability
Model
Telecommunication links

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

@article{4202e31b8aae45b2a579e8b459415114,
title = "Analytic models of loss recovery of TCP Reno with packet losses",
abstract = "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.",
author = "Beomjoon Kim and Lee, {Jai Yong}",
year = "2003",
month = "12",
day = "1",
language = "English",
volume = "2662",
pages = "938--947",
journal = "Lecture Notes in Computer Science",
issn = "0302-9743",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - Analytic models of loss recovery of TCP Reno with packet losses

AU - Kim, Beomjoon

AU - Lee, Jai Yong

PY - 2003/12/1

Y1 - 2003/12/1

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

M3 - Article

AN - SCOPUS:21144432930

VL - 2662

SP - 938

EP - 947

JO - Lecture Notes in Computer Science

JF - Lecture Notes in Computer Science

SN - 0302-9743

ER -