2008
DOI: 10.1145/1452335.1452338
|View full text |Cite
|
Sign up to set email alerts
|

Removing exponential backoff from TCP

Abstract: The well-accepted wisdom is that TCP's exponential backoff mechanism, introduced by Jacobson 20 years ago, is essential for preserving the stability of the Internet. In this paper, we show that removing exponential backoff from TCP altogether can be done without inducing any stability sideeffects. We introduce the implicit packet conservation principle and show that as long as the endpoints uphold this principle, they can only improve their end-to-end performance relative to the exponential backoff case.By con… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
14
0

Year Published

2010
2010
2021
2021

Publication Types

Select...
4
3
1

Relationship

0
8

Authors

Journals

citations
Cited by 27 publications
(14 citation statements)
references
References 40 publications
0
14
0
Order By: Relevance
“…Significant contention in a wireless mesh often leads to high packet loss, causing TCP to go into exponential backoff [31] and to starve. We found that we can prevent TCP starvation if we retransmit failed packets up to three more times using a second hardware queue, even after 802.11 gives up.…”
Section: Optimizationsmentioning
confidence: 99%
“…Significant contention in a wireless mesh often leads to high packet loss, causing TCP to go into exponential backoff [31] and to starve. We found that we can prevent TCP starvation if we retransmit failed packets up to three more times using a second hardware queue, even after 802.11 gives up.…”
Section: Optimizationsmentioning
confidence: 99%
“…When a timeout occurs, the sender retransmits the lost segment and backs off by doubling the RTO. This exponential backoff was introduced as a counter-measure to congestion collapses in the early 1980s (although, some researchers claim that the RTO backoff could safely be removed from TCP without causing instability to the Internet [118]). Algorithm 2 outlines the basic procedure for computing the RTO as specified in RFC 2988 [128].…”
Section: Timeout-based Retransmissionmentioning
confidence: 99%
“…the median connection duration was around 1 second and after this point the median duration falls to 100-200 msec. The short duration of connections suggests that seemingly small changes to the delivery process that save modest amounts of wall-clock time may ultimately benefit the user experience more than one might think at first blushe.g., Early Retransmit [2] and reducing TCP's traditional exponential backoff between retransmissions [10]. Figure 6 also illustrates the time between establishing a connection and sending an HTTP request.…”
Section: Http Transaction Characterizationmentioning
confidence: 99%