Fairness and stability of congestion control mechanisms of TCP

Hasegawa, Go; Murata, Masayuki; Miyahara, Hideo

doi:10.1109/infcom.1999.752151

Cited by 65 publications

(50 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…If RTTs become small, on the other hand, TCP Vegas determines that the network is relieved from the congestion, and increases cwnd. In congestion avoidance phase, the cwnd is updated as shown in [6]:…”

Section: Congestion Avoidance Mechanismmentioning

confidence: 99%

Comparison of TCP congestion control mechanisms Tahoe, Newreno and Vegas

Kumpavat¹

2013

IOSR-JECE

View full text Add to dashboard Cite

Section: Congestion Avoidance Mechanismmentioning

confidence: 99%

Comparison of TCP congestion control mechanisms Tahoe, Newreno and Vegas

Kumpavat¹

2013

IOSR-JECE

View full text Add to dashboard Cite

“…In particular two major limitations of other TCP models are overcome in our model: 1. Our model is the first model of TCP-Vegas sources sending on-off traffic; all previous work on modeling TCP-Vegas only allowed bulk transfers [1,2,9,12,13,16,19]. Bulk transfer models are usually associated with FTP traffic, where…”

mentioning

confidence: 99%

Modeling TCP-vegas under on/off traffic

Wierman

Osogami

Olsén

2003

SIGMETRICS Perform. Eval. Rev.

View full text Add to dashboard Cite

There has been a significant amount of research toward modeling variants of the Transmission Control Protocol (TCP) in order to understand the impact of this protocol on file transmission times and network utilization. Analytical models have emerged as a way to reduce the time required for evaluation when compared with more traditional evaluations performed using event driven simulators such as ns. In addition, when designed carefully, analytical models help researchers make design decisions about novel TCP mechanisms.A wide variety of techniques have been applied to the problem of TCP modeling with a fair amount of success. These techniques range over renewal theory [10,17], Markov chains [11], and fluid models [1]. However, until the recent seminal work of papers using fix-point methods [3,4,5,6,7,8,14,15,18], no modeling technique was able to mimic both the structure of a TCP source and the interaction a source has with the network. The fix-point methods take the novel approach of separating the modeling of network behavior from the modeling of the behavior within a TCP source, and then allowing the two to tune each other via feedback. The fix-point framework allows general network topologies to be analyzed, and issues such as the interpretation of end-to-end loss rates in multiple bottleneck networks are addressed in [3,7,8,15,18].In this paper, we generalize the framework based on a fixed point method introduced by Casetti and Meo in [4,5] in order to allows us to model TCP-Vegas connections. The framework of Casetti and Meo, which uses a Markov chain to model the TCP source, has a few advantages over other fixed point methods: (1) it can model explicit details of TCP, making it possible to distinguish different flavors of TCP; (2) it allows modeling on-off traffic sources; (3) it gives the fraction of time that TCP spends in each state, from which we can evaluate the effectiveness of each mechanism of the protocol.A majority of existing analytical models focus on TCP-Reno, the most widely deployed variant of TCP, and there has been little research on analytical models of TCP-Vegas, a more recently proposed variant. Analytical models of TCP-Vegas have been difficult to develop because TCP-Vegas uses observed delay to detect an incipient stage of congestion and try to adjust the sending rate before packets are lost. Prior studies on measurement and simulation of the performance of TCPVegas suggest that in many situations it is able to provide users higher throughput and lower loss rates than TCP-Reno.Hence, it is an important task to model the performance of TCP-Vegas in order to understand how this protocol performs in a network shared with other variants of TCP.Because TCP-Vegas uses observed delay, as well as loss events, to adjust the congestion window size, while TCP-Reno uses only loss events, the extension made in this work to the framework of Casetti and Meo for modeling TCP-Reno is nontrivial. We need to make two major changes. First, analyzing the network model to determine the loss rate and the...

show abstract

“…Prior research demonstrates that Vegas (in isolation) generally performs better than other implementations of TCP [1,3,4,9,13]. Ahn et al [1] and Mo et al [13] also show that when a Vegas connection competes with a Reno connection, Vegas does not receive a fair share of bandwidth due to its conservative congestion-avoidance mechanism.…”

Section: Compatibility Of Tcp Reno and Tcp Vegasmentioning

confidence: 99%

“…First, setting « ¬ introduces stability problems [2], i.e., the congestion window oscillates around the equilibrium value. Second, setting « and ¬ too far apart creates a larger stability region than needed, resulting in connections that can converge to opposite ends of the stability region, thus affecting fairness [2,9]. …”

Section: Approachmentioning

confidence: 99%

“…Further, recent work shows that this oscillatory behavior induces chaotic behavior into the network [4,15], thus adversely affecting overall network performance. In contrast, TCP Vegas generally performs better with respect to overall network utilization [1,3], stability [9,13], fairness [9,13], throughput and packet loss [1,3], and burstiness [4] when the entire network consists of Vegas-only connections. However, research to date has shown that when Reno and Vegas perform head-to-head, Reno generally steals bandwidth from Vegas [1,13].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation