Y. Richard Yang, Min Sik Kim, and Simon S. Lam, "Transient behavior of TCP-friendly congestion control protocols," in Proceedings of INFOCOM, April 2001.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and ANI 9986397. Shudong Jin was also supported by an IBM PhD Research Fellowship. relationship 1= R p) should hold, where is the throughput of a ow, p is the loss rate, and R is the round trip time. In addition to TCP friendliness, smoothness, aggressiveness, and responsiveness [1, 9] are important indices of congestion control performance. Smoothness indicates the variability in transmission rate. Aggressiveness indicates how fast a connection probes extra bandwidth by opening up its window. Responsiveness measures how fast a connection reacts to increased congestion by ....

....decreasing its window size. Smoothness characterizes the steady state behavior of congestion control protocols, whereas both aggressiveness and responsiveness characterize transient behavior. An important observation is that there are tradeo s among smoothness, aggressiveness, and responsiveness [1, 9]. Comparisons of TCP, general AIMD [1, 3] TFRC [5] and TEAR [2] have shown that typically higher smoothness means less aggressiveness and responsiveness. Several questions remain unanswered. First, both window based and equation based congestion control schemes have been studied recently. ....

[Article contains additional citation context not shown here]

Y. Richard Yang, Min Sik Kim, and Simon S. Lam, \Transient behavior of TCP-friendly congestion control protocols," in Proceedings of INFOCOM, April 2001.

....modifying the OMQN structure. 3 TCP Protocol Modification The basic congestion avoidance algorithm of TCP NewReno (presently the most diffused TCP version) is AIMD (Additive Increase Multiplicative Decrease) The basic properties of AIMD algorithms, generalizing the TCP protocol, were studied in [22, 23]. Several approximations are introduced for analytical tractability, the main of which is the incorrelation of the loss process with the TCP protocol, which does not allow to fully appreciate any property of the protocol that leads to a smaller loss probability. Recently, studies like [12, 15] ....

Y. Yang, M. Kim, S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols, " IEEE Infocom'01, Anchorage, AK, USA April 22--26, 2001.

....such fairness, one solution is to satisfy TCP friendliness, which means the (#, p) relationship # 1 (R # p) should hold, where # is the throughput of a flow, p is the loss rate, and R is the round trip time. In addition to TCP friendliness, smoothness, aggressiveness, and responsiveness [1, 9] are important indices of congestion control performance. Smoothness indicates the variability in transmission rate. Aggressiveness indicates how fast a connection probes extra bandwidth by opening up its window. Responsiveness measures how fast a connection reacts to increased congestion by ....

....decreasing its window size. Smoothness characterizes the steady state behavior of congestion control protocols, whereas both aggressiveness and responsiveness characterize transient behavior. An important observation is that there are tradeoffs among smoothness, aggressiveness, and responsiveness [1, 9]. Comparisons of TCP, general AIMD [1, 3] TFRC [5] and TEAR [2] have shown that typically higher smoothness means less aggressiveness and responsiveness. 1.1 Motivation Our work is motivated by the need for new controls that have high smoothness in steady state and high aggressiveness ....

[Article contains additional citation context not shown here]

Y. Richard Yang, Min Sik Kim, and Simon S. Lam, "Transient behavior of TCP-friendly congestion control protocols," in Proceedings of IEEE INFOCOM, April 2001.

....modifying the OMQN structure. 3 TCP Protocol Modification The basic congestion avoidance algorithm of TCP NewReno (presently the most diffused TCP version) is AIMD (Additive Increase Multiplicative Decrease) The basic properties of AIMD algorithms, generalizing the TCP protocol, were studied in [22, 23]. Several approximations are introduced for analytical tractability, the main of which is the incorrelation of the loss process with the TCP protocol, which does not allow to fully appreciate any property of the protocol that leads to a smaller loss probability. Recently, studies like [12, 15] ....

Y. Yang, M. Kim, S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols, " IEEE Infocom'01, Anchorage, AK, USA April 22--26, 2001.

....throughput happens to be for positive autocorrelation of the loss event inter arrival times (not shown here due to the space limitations) 5. Discussion and conclusion It was observed through empirical studies that TFRC is overly conservative for high loss event ratios (e.g. see [12]) however, no explanation is provided. Our observations made in the present paper suggest an answer to this. It is due to randomness of the loss event ratio estimator n, and non linearity involved due to both updating the rate at the loss events and non linearity of the function f. We support ....

Y. Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of TCPfriendly congestion control protocols. In Proc. of IEEE Infocorn'2001, March 2001.

....from two different perspectives, namely the application requirements and the characteristics of the underlying networks. The former expounds the impact of the transmission gaps caused by halving the transmission rate during congestion on the quality of delay sensitive applications; authors in [4, 5, 8, 9] propose TCP friendly protocols that satisfy two fundamental goals: i) to achieve smooth window adjustments; this is done by reducing the window decrease ratio during congestion, and (ii) to compete fairly with TCP flows; this is approached by reducing the window increase factor according to a ....

....link. The number of flows N increases with time as follows: # # # # # # # # # # # # # # # sec) 25 20 ( 50 sec) 20 15 ( 25 sec) 15 0 ( 12 t t t t That is, N doubles every 5 seconds after a 15 second period. Note that the metrics now are allotted fairness, as shown in Figure 6. [9] claims that generating smoother traffic improves allotted fairness, as confirmed in the time period from t 0 =0 to t 1 =15 in Figure 6. However, when new flows join after 15 seconds, fairness displays a dependency on the sending rate of existing and incoming flows. In this context, the responsive ....

Y.R. Yang, M.S. Kim and S.S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols", Proceedings of IEEE INFOCOM 2001.

....and transition probabilities, rather than modifying the OMQN structure. 3 TCP Protocol Modification The basic congestion avoidance algorithm of TCP NewReno (presently the most diffused TCP version) is AIMD. The basic properties of AIMD algorithms, generalizing the TCP protocol, were studied in [22, 23]. Several approximations are introduced for analytical tractability, the main of which is the incorrelation of the loss process with the TCP protocol, which does not allow to fully appreciate any property of the protocol that leads to a smaller loss probability. Recently, studies like [12, 15] ....

Y. Yang, M. Kim, S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols. " IEEE Infocom'01, Anchorage, AK, USA April 2001.

....[24] 3 Multiplicative decrease causes transmission gaps that hurt the performance of real time applications that experience jitter and degraded goodput. Furthermore, multiplicative decrease with a factor of or a window adjustment to 2 packets characterizes a rather conservative strategy [9, 13, 16, 24, 30, 31]. 4 Error detection lacks an appropriate classification module that would permit a responsive strategy, oriented by the nature of potential errors. That is, when errors appear to be transient due to short lived flows or random wireless interference, congestion control mechanisms (i.e. timeout ....

Y.R. Yang, M.S. Kim and S.S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols", In Proceedings of IEEE INFOCOM 2001, April 2001.

....I and R D guarantee convergence to fairness. In the context of I D congestion control, convergence to fairness is usually defined as the ability of any number of identical flows sharing a common bottleneck Sometimes called TCP compatible [4] 8] or TCP friendly [21] Note that some papers [4] [20], 21] use a different notation, in which WD (w) 1 #)w and this formula has a different form. Furthermore, if the rate of AIMD is dominated by timeouts, the formula assumes yet another form [21] 0 7803 7476 2 02 17.00 (c) 2002 IEEE. link to reach a state in which their rates become ....

Y.R. Yang, M.S. Kim, and S.S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols," IEEE INFOCOM, April 2001.

....measures how fast the connection reacts to increased congestion by decreasing its window size. It is desirable that the connection reduces its transmission rate to its fair share promptly. Both aggressiveness and responsiveness are measures of the transient behavior of congestion control protocols [21]. Convergence measures how fast competing connections converge to their fair share of bandwidth. Convergence speed is related to the aggressiveness and responsiveness indices. More aggressive and responsive protocols usually converge faster. Several recently proposed TCP friendly congestion ....

....connections converge to their fair share of bandwidth. Convergence speed is related to the aggressiveness and responsiveness indices. More aggressive and responsive protocols usually converge faster. Several recently proposed TCP friendly congestion control schemes, including general AIMD [8, 21], binomial algorithms [1] TFRC [9] and TEAR [19] can provide smoother transmission rate than TCP. However, these algorithms may lack the aggressiveness and responsiveness of TCP. Hence, when network conditions change drastically, these protocols can not react to the change promptly. In ....

[Article contains additional citation context not shown here]

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behavior of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM, April 2001.

.... In addition to characterizing the sending rate of these SlowCC algorithms given a steady state packet loss rate, there has been some exploration of the relative fairness of these mechanisms with TCP in the presence of ON OFF background traffic and under different levels of statistical multiplexing [6, 19]. In addition, the relative smoothness of SlowCC proposals has been explored, with several different metrics used to measure smoothness [8, 19] There has also been some investigation of the effect of SlowCC proposals on queue dynamics, including the effect on oscillations in the queue size, both ....

.... of the relative fairness of these mechanisms with TCP in the presence of ON OFF background traffic and under different levels of statistical multiplexing [6, 19] In addition, the relative smoothness of SlowCC proposals has been explored, with several different metrics used to measure smoothness [8, 19]. There has also been some investigation of the effect of SlowCC proposals on queue dynamics, including the effect on oscillations in the queue size, both with and without active queue management [7] While there has been a preliminary evaluation of some SlowCC algorithms to measure aggressiveness ....

[Article contains additional citation context not shown here]

YANG, Y., KIM, M., AND LAM, S. Transient Behaviors of TCP-friendly Congestion Control Protocols. In Proceedings of the Conference on Computer Communications (IEEE Infocom) (Anchorage, AK, April 2001), pp. 1716--1725.

.... for our JSNM protocol to behave fairly with respect to coexistent TCP flows and degrade gracefully to TCP like behavior under unfavorable conditions (i.e. conditions not designed for JSNM applications) 64] A detailed survey on TCP friendly congestion control algorithms can be found in [65] and [66]. In this thesis, we propose a TCP friendly congestion control mechanism roughly based on the original TCP congestion control algorithm. It mimics the long term behavior of TCP and yet does not have the saw tooth pattern typical of TCP tra#c, which introduces large variation in received video ....

Y. R. Yang, S. K. Min, and S. S. Larn, "Transient behaviors of TCP-friendly congestion control protocols," in Proceedings of INFOCOM 2001, vol. 3, pp. 1716--1725. 124

....measures how fast the connection reacts to increased congestion by decreasing its window size. It is desirable that the connection reduces its transmission rate to its fair share promptly. Both aggressiveness and responsiveness are measures of the transient behavior of congestion control protocols [21]. Convergence measures how fast competing connections converge to their fair share of bandwidth. Convergence speed is related to the aggressiveness and responsiveness indices. More aggressive and responsive protocols usually converge faster. Several recently proposed TCP friendly congestion ....

....connections converge to their fair share of bandwidth. Convergence speed is related to the aggressiveness and responsiveness indices. More aggressive and responsive protocols usually converge faster. Several recently proposed TCP friendly congestion control schemes, including general AIMD [8, 21], binomial algorithms [1] TFRC [9] and TEAR [19] can provide smoother transmission rate than TCP. However, these algorithms may lack the aggressiveness and responsiveness of TCP. Hence, when network conditions change drastically, these protocols can not react to the change promptly. In ....

[Article contains additional citation context not shown here]

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behavior of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM, April 2001.

.... In addition to characterizing the sending rate of these SlowCC algorithms given a steady state packet loss rate, there has been some exploration of the relative fairness of these mechanisms with TCP in the presence of ON OFF background traffic and under different levels of statistical multiplexing [6, 19]. In addition, the relative smoothness of SlowCC proposals has been explored, with several different metrics used to measure smoothness [8, 19] There has also been some investigation of the effect of SlowCC proposals on queue dynamics, including the effect on oscillations in the queue size, both ....

.... of the relative fairness of these mechanisms with TCP in the presence of ON OFF background traffic and under different levels of statistical multiplexing [6, 19] In addition, the relative smoothness of SlowCC proposals has been explored, with several different metrics used to measure smoothness [8, 19]. There has also been some investigation of the effect of SlowCC proposals on queue dynamics, including the effect on oscillations in the queue size, both with and without active queue management [7] While there has been a preliminary evaluation of some SlowCC algorithms to measure aggressiveness ....

[Article contains additional citation context not shown here]

YANG, Y., KIM, M., AND LAM, S. Transient Behaviors of TCP-friendly Congestion Control Protocols. In Proceedings of the Conference on Computer Communications (IEEE Infocom) (Anchorage, AK, April 2001), pp. 1716--1725.

....probes available bandwidth by increasing its congestion window size linearly, and responds to increased congestion (indicated by packet losses) by decreasing the window size multiplicatively. Recently proposed congestion control mechanisms include generalizations of TCP like window based schemes [3, 11, 26, 30] and equation based schemes [12, 23, 29] One common objective of these new schemes is to reduce variations in transmission rate. Such high variations may limit network utilization. In addition, they are not desirable for emerging applications such as real time streaming applications on the ....

....means how fast the connection reacts to increased congestion and decreases its window size. It is desirable that the connection reduces its transmission rate to its fair share promptly. Both aggressiveness and responsiveness are measures of the transient behavior of congestion control protocols [30]. Convergence means whether and how fast competing connections converge to their fair share of bandwidth. Certainly, convergence speed is related to the aggressiveness and responsiveness indices. More aggressive and responsive protocols usually converge faster. Several recently proposed ....

[Article contains additional citation context not shown here]

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behavior of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM, April 2001.

No context found.

Yang, Y., Kim, M., and Lam, S. 2001. Transient behaviors of tcp-friendly congestion control protocols. In Proceedings of the IEEE Infocom. IEEE, Anchorage, AK.

No context found.

Y. Yang, M. Kim, and S. Lam. Transient behaviors of tcpfriendly congestion control protocols. In Proceedings of the IEEE Infocom, Anchorage, AK, Apr. 2001. IEEE.

No context found.

Yang, Y., Kim, M., Lam, S.: Transient Behaviors of TCP-friendly Congestion Control Protocols. IEEE Infocom 2001, Anchorage, AK (2001)

No context found.

Yang, Y., Kim, M., Lam, S.: Transient Behaviors of TCP-friendly Congestion Control Protocols. IEEE Infocom 2001, Anchorage, AK (2001)

No context found.

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCP-friendly congestion control protocols. Computer Networks, 41(2):193--210, Feb. 2003.

No context found.

Yang Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of TCP-friendly congestion control protocols. Technical Report TR--00--23, Department of Computer Sciences, The University of Texas, Austin, Texas, U.S.A., September 2000. 34

No context found.

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCPfriendly congestion control protocols. In Proceedings of IEEE INFOCOM 2001.

No context found.

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM 2001.

No context found.

Y. Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of TCPfriendly congestion control protocols. Computer Networks, 41(2):193--210, February 2003.

....is plotted in Figure 8. The ratio of the average incoming rates decreases linearly as CoV increases. In order to achieve a ratio higher than 0.8, CoV should not exceed 0.3. Some congestion control protocols designed to avoid sending rate fluctuations have sending rate CoV lower than 0. 3 [18]; therefore, the throughput of one of these protocols would be suitable for edge bandwidth estimation in our algorithm implementation. Protocols with larger CoV like the AIMD (additive increase multiplicative decrease) protocol of TCP can also be used by having su#ciently large measurement ....

Y. Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM 2001.

....a continuum of protocols with di erent degrees of smoothness. While smoother protocols are better from the application s point of view, this is true only in steady state. Aggressive protocols are also more responsive to congestion indications and thus have better transient and dynamic behaviors [3, 29]. We design and experimentally investigate a 1 CYRF protocol called LOG, with aggressiveness and smoothness properties intermediate to that of IIAD and SQRT, the two other non linear (binomial) congestion controls that have been previously studied [2] Finally, we brie y describe a CYRF protocol ....

....We will use these results to design a smooth, TCP friendly CYRF protocol called LOG. 5.1 Smoothness Intuitively, the smoother a ow is, the lesser will be its decrease in response to a congestion indication from the network. Smoothness has been used as a metric in previous work (e.g. 9] and [29]) In this work, we say a protocol is smooth if its increase and decrease policies are smooth. An window increase or decrease policy x t R x t x, is said to be smooth if j xj x (16) Suppose the increase policy of a smooth protocol is successively applied two times: x t R x t x t , x ....

[Article contains additional citation context not shown here]

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCPfriendly congestion control protocols. In Proceedings of IEEE INFOCOM 2001.

....[15] in window based approaches [16] However, 3] which also shows the importance of self clocking in dynamic situations, gives an ad hoc mechanism to limit the TFRC sending rate following a loss event, that performs well in the test scenarios considered. Floyd et al. 11] Yang et al. [40] and Bansal et al. 3] compare the performance of several window based and rate based protocol under various dynamic or transient conditions. This issue is not explored in our work. An important distinction between CYRF and other approaches (including other window based approaches) is the wide ....

....indication from the network. Smoothness has been variously measured by the largest reduction of the sending rate in one round trip time in the deterministic steady state scenario (Floyd et al. 11] and the coecient of variation of the time series representing successive window sizes (Yang et al. [40]) In this work, we look at smoothness as an (optional) characteristic of a WBF protocol rather than as a metric. We say a WBF protocol is smooth if its increase and decrease policies are smooth. An window increase or decrease policy P : x t R x t x, where x is the window size, is said to be ....

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCP-friendly congestion control protocols. In Proceedings of IEEE INFOCOM

....B. Analytical results We present our analytical results on ### #### for TCP, GAIMD, TFRC, and TEAR. The derivations of these results are nontrivial and require several pages to present. They are omitted herein due to page limitation. The derivations are presented in our technical report [23]. B.1 AIMD At low loss rate, assuming Poisson loss arrival, we derive ### #### for AIMD (including GAIMD and TCP Reno as special cases) to be: ### #### #### # # # # # ### (4) where # is the reduction ratio of congestion window size when there is a congestion indication. Plugging # ....

....Poisson loss, we derive ### #### for TEAR to be: ### #### #### # #### (8) B.3 TFRC At low loss rate, assuming Bernoulli loss, we derive ### #### for TFRC to be: ### #### #### # #### (9) At high loss rate (about 20 ) we derive that ### #### for TFRC will be between 0.8 and 2. 4 [23]. C. Simulation results C.1 High multiplexing environments We start our simulation with periodic loss. Figure 3 shows flow sending rate traces when the loss rate is 5 . For this fig0 10000 20000 30000 40000 50000 60000 70000 80000 500 550 600 650 700 Rate (bytes second) RTT TCP ....

[Article contains additional citation context not shown here]

Yang Richard Yang, Min Sik Kim, and Simon S. Lam, "Transient behaviors of TCP-friendly congestion control protocols," Tech. Rep. TR--00-- 23, Department of Computer Sciences, The University of Texas, Austin, Texas, U.S.A., Sept. 2000.

No context found.

Y. Richard Yang, Min Sik Kim, and Simon S. Lam, "Transient behavior of TCP-friendly congestion control protocols," in Proceedings of INFOCOM, April 2001.

No context found.

Yang Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of TCPfriendly congestion control protocols. In Proceedings of INFOCOMM 2001.

No context found.

Yang Richard Yang, Min Sik Kim, and Simon S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols",Proceedings of IEEE Infocom 2001.

No context found.

Y. R. Yang, M. S. Kim, and S. S. Lam, "Transient behaviors of TCP-friendly congestion control protocols," The International Journal of Computer and Telecommunications Networking, vol. 41, pp. 193--210, Feb. 2003.

No context found.

Y.R. Yang, M.S. Kim, and S.S. Lam, "Transient Behaviors of TCPfriendly Congestion Control Protocols," IEEE INFOCOM, April 2001.

No context found.

Y. Yang, M. Kim, and S. Lam. Transient Behaviors of TCP-friendly Congestion Control Protocols, Proceedings of INFOCOM 2001.

No context found.

Y. Yang, M. Kim, and S. S. Lam, "Transient behaviors of tcpfriendly congestion control protocols," in Proc. IEEE INFOCOM 2001.

No context found.

Y.R. Yang, M.S. Kim and S.S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols", Proceedings of IEEE INFOCOM 2001.

No context found.

Y. R. Yang, M. Kim, and S. Lam, "Transient behaviors of TCP-friendly congestion control protocols," in Proceedings of the Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM-01). Los Alamitos, CA: IEEE Computer Society, Apr. 22--26 2001, pp. 1716--1725.

No context found.

Y. Richard Yang, Min Sik Kim, and Simon S. Lam. Transient Behaviors of TCP-friendly Congestion Control Protocols. In Proc. of IEEE Infocom'2001.

No context found.

Y. Yang, M. Kim, and S. Lam, "Transient behaviors of TCP-friendly congestion control protocols," In Proceedings of IEEE INFOCOM 2001, Anchorage, Alaska, Apr. 2001.

No context found.

Y.R. Yang, M.S. Kim, and S.S. Lam, "Transient Behaviors of TCPfriendly Congestion Control Protocols," IEEE INFOCOM, April 2001.

No context found.

Y. Yang, M. Kim, and S. S. Lam. Transient behaviors of tcpfriendly congestion control protocols. In Proc. IEEE INFOCOM 2001.

No context found.

Y. Yang, M. Kim, and S. Lam, "Transient Behaviors of TCP-friendly Congestion Control Protocols," IEEE INFOCOM, April 2001.

No context found.

Y. R. Yang, M. S. Kim, and S. S. Lam. Transient behaviors of TCP-friendly congestion control protocols. In Proc. of IEEE INFOCOM'01, Apr. 2001.

No context found.

Y. Richard Yang, Min Sik Kim, and Simon S. Lam. Transient behaviors of tcp-friendly congestion control protocols. In Proc. of IEEE Infocom'2001.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC