Wang, Y.T., and B. Sengupta, "Performance analysis of a feedback congestion control policy under non-negligible propagation delay," Proc of ACM Sigcomm, pp. 149-157, Zurich, September 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Any such scheme can, therefore, at best be a congestionavoidance algorithm. Care must be taken to ensure that transmitters are not overly pessimistic. 8. 4 Congestion Control: Analysis and Simulation Several performance studies on congestion control algorithms have been reported re cently [2, 11, 26, 25, 30, 38, 40]. Several of them have reported that feedback delay plays an important role in stability and fairness. Most authors, however, have assumed a fixed (deterministic) delay because the actual distribution was not known. One contribution of this paper is an empirical observation on the distribution of ....

Wang, Y.T., and B. Sengupta, "Performance analysis of a feedback congestion control policy under non-negligible propagation delay," Proc of ACM Sigcomm, pp. 149-157, Zurich, September 1991.

....Since the amount of bandwidth mismatch depends on the propagation delay, increased propagation delay reduces the effectiveness of feedback based flow control schemes. The effects of increased propagation delay on the effectiveness of feedback based flow control schemes have been studied in past [14], 15] 16] 17] But, in addition to the propagation delay, some other system parameters may have considerable effects on the feedback based flow control schemes. In this paper, the impact of these system parameters (network, VBR and ABR time scales) will be investigated. t 1 t 2 t 1 2T d t 2 ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under non-negligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

.... Several studies have shown that adaptive rate control schemes are adversely affected by the presence of propagation delays between the controlled sources and the control nodes [18] and several authors have limited the applicability of such schemes to the realm of local area networks [19] [20] where propagation delays are no longer than several tens of the service time. Unlike in open loop schemes where no competition for network bandwidth takes place once a call is set up, an important characteristic of adaptive rate control schemes is that existing connections compete, at the cell ....

....difficulty, most previous analytical studies for adaptive rate control schemes have focused on aggregate performance measures. In addition to the large bandwidth propagation delay products characterizing high speed networks, an often overlooked by earlier discrete time studies such as [20], 14] characteristic is the potentially large disparity between the relative speeds of a network link and an input source. This disparity or speed up factor represents the time in slots required for a source to generate enough bits to form one cell and is referred to as the source time ....

[Article contains additional citation context not shown here]

Y. T. Wang and B. Sengupta. Performance analysis of a feedback congestion control policy under non-negligible propagation delay. In Sigcomm'91, pages 149--157, Zurich, Switzerland, Sep 1991.

....occurs which may induce cell losses depending on the amount of buffering available. When an increase in the remaining bandwidth occurs, underutilization occurs for a duration of 2T d . The effects of increased propagation delay on the flow control schemes has been studied in the past [11], 12] 13] It turns out that, as the distance between the sources and the network access node increases (propagation delay increases) more cell losses occur. t 1 t 2 t 1 2T d t 2 2T d rate time total rate at the network access node ABR rate VBR CBR rate Figure 1: An example of ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under non-negligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

....to get achieved. Recently, the advent of broadband packet networks has raised a special set of issues induced by the large increase in the propagation delay bandwidth product as well as the speed mismatch between the increase in data transfer rate and limited processing power at routing stations [5, 6, 7, 11, 14, 31]. A general consensus, albeit tentative, has taken shape which recommends that congestion control protocols be simplified to minimize switching overhead with a resultant emphasis on end to end protocols, and proactive schemes are to be preferred over reactive ones due to the high cost associated ....

Y. T. Wang and B. Sengupta. Performance analysis of a feedback congestion control policy under non-negligible propagation delay. In Proc. ACM SIGCOMM '91, pages 149--157, 1991. 24

....However, due to limited space we do not describe the numerical results in this paper. Such a description is available in [21] and [22] Considerable work has been done recently by several authors on the use of delayed feedback for flow control in high speed networks (see [17] 18] 8] 19] [26], 23] 25] 14] A study of feedback in congestion control in ATM networks has to incorporate both the dynamics of the traffic sources and the relatively large propagation delay. In [17] 18] 8] 19] 26] the authors study the effect of large propagation delays on feedback mechanisms. ....

....delayed feedback for flow control in high speed networks (see [17] 18] 8] 19] 26] 23] 25] 14] A study of feedback in congestion control in ATM networks has to incorporate both the dynamics of the traffic sources and the relatively large propagation delay. In [17] 18] 8] 19] [26], the authors study the effect of large propagation delays on feedback mechanisms. These papers provide much insight into the difficulties of using delayed feedback for flow control in high speed data networks. Feedback mechanisms for ABR mechanisms have also been studied extensively (see ....

Y. T. Wang and B. Sengupta. Performance analysis of a feedback congestion control policy under non-negligible propogation delay. In Proc. ACM SIGCOMM, 1991.

....is to find an optimal control policy for the ABR source that balances various performance measures such as throughput, mean and variance of delay and probability of overflow. Often this problem is addressed by minimizing the distance of stationary mean queue length from a given fixed constant N 0 [34], 18] 30] In this paper, we consider parametrized policies that have several levels of control. We develop a simultaneous perturbation stochastic approximation (SPSA) 33] variant of a two timescale stochastic approximation algorithm [6] to obtain the optimal policy with this structure (see ....

....for ABR flow control. As mentioned earlier, flow control in ABR service requires balancing various conflicting performance criteria such as mean and variance of delay and throughput. Often this is addressed by minimizing the distance of stationary mean queue length from a given fixed constant N 0 [34], 18] 30] 5] We adopt a similar approach here, i.e. h(x) x N 0 , where N 0 is assumed given. In the concluding section, we shall also indicate ways to obtain an optimal such N 0 . We compare the performance of optimal structured closed loop feedback policies of type (2.2) obtained by ....

Y. T. Wang and B. Sengupta. Performance analysis of a feed back congestion control policy. ACM SIGCOMM Computer Communication, pages 149--157, 1991.

....mismatch cycles and the magnitude of bandwidth mismatch is proportional to the propagation delay, the effectiveness of a rate based flow control scheme decreases with increased propagation delay. The impact of increased propagation delay on the flow control schemes has been studied in the past [9], 10] 11] 12] Besides the physical distance other system parameters namely the network, ABR and VBR time scales impact considerably on the effectiveness of the flow control schemes. In this paper, the impact of these system parameters on the effectiveness of the feedback based flow control ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under nonnegligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

....to temporarily stop sending any new frames until the congestion has passed. This switch flow control mechanism is built into the input port. It uses a combination of XOFF XON thresholds and XOFF XON control frames to ensure that no data frames are dropped due to a lack of free memory in the switch [Wan91]. The basic idea of the backpressure flow control is shown in Figure 2. When an outgoing link of a switch becomes congested, frames coming into this switch cannot be sent via the congested link, and hence they need to be stored inside the switch. If this situation persists the storage within the ....

.... (FIFO Head of FIFO queue Input Port FIFO queue filling emptying current FIFO queue occupancy low water mark XON threshold high water mark XOFF threshold send XOFF frame send XON frame Figure 2: Basics of the switch flow control mechanism queue) with the XOFF threshold value (high water mark [Wan91]) If the XOFF threshold is reached, an XOFF control frame is sent to the adjacent switch (or the source) which in turn stops sending any new data frames. This causes the FIFO queue to release data until the current queue occupancy reaches the XON threshold (low water mark) The flow of data ....

Wang Y. and Sengupta B. Performance Analysis of a Feedback Congestion Control Policy Under NonNegligable Propagation Delay, ACM Sigcomm 1991

....Most of the previous analysis assume lossless systems with the buffer overflow probabilities used to estimate the loss probabilities. The related works along these lines may be found in [6] 7] 8] and [9] In reality, the traffic sources are stochastic and so too are the feedback delays. In [10], the authors model a simplified delayed feedback system with constant service rate and independent random arrivals with on off source behaviour. In [11] and [12] the authors analyze a stochastic feedback system using Markov modulated fluid models for the traffic sources, a fluid queue and ....

....may be obtained to any degree of accuracy. This analysis can be extended to control based on multiple queue thresholds. We may wish to explore the possibility of two level overload control for the feedback system. While this hysteresial scheme has been proposed by several researchers ( 9] [10]) its performance has not been analyzed in the context of ABR traffic. In [10] the authors present some analysis on the two level control scheme in a much simplified model of a feedback system. They give some insight into the choice of the two thresholds for this simplified model to optimize the ....

[Article contains additional citation context not shown here]

Y. T. Wang and B. Sengupta, "Performance Analysis of a feedback Congestion Control policy under nonnegligible propogation delay," Proc. ACM SIGCOMM,91, 1991.

....occurs which may induce cell losses depending on the amount of buffering available. When an increase in the remaining bandwidth occurs, underutilization occurs for a duration of 2T d . The effects of increased propagation delay on the flow control schemes has been studied in the past [11], 12] 13] It turns out that, as the distance between the sources and the network access node increases (propagation delay increases) more cell losses occur. t 1 t 2 t 1 2T d t 2 2T d rate time overutilization underutilization total rate at the network access node ABR rate VBR CBR rate ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under non-negligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

....outdated information. Furthermore, the effect of a rate adjustment at a source is felt at the bottleneck node only after another delay. These problems have led many researchers to investigate the impact of the feedback and control delays on the performance of flow and congestion control mechanisms [1, 4, 6, 10, 20, 21, 26]. It is well known in control theory that the existence of delays degrades the performance of conventional feedback control mechanisms. Specifically, a time delay adds a phase shift and reduces the margin of stability of a closed loop system. Stability can be preserved by reducing the controller ....

....shift and reduces the margin of stability of a closed loop system. Stability can be preserved by reducing the controller gain, however at the expense of the transient response [19] Recent analyses of flow control mechanisms in networks with non negligible delays have led to similar conclusions [26, 4, 10]. The existence of time delays also gives rise to oscillatory and other complicated behavior. Large amplitude periodic and chaotic oscillations, and their adverse effect on bandwidth utilization, have indeed been observed in a wide range of communication networks [17, 9, 27] These difficulties ....

[Article contains additional citation context not shown here]

Y-T. Wang, B. Sengupta, "Performance analysis of a feedback congestion control policy under non-negligible propagation delay", Proc. ACM SIGCOMM `91, Zurich, Switzerland, Sept. 1991, pp. 149-158.

....associated with feedback congestion control. A taxonomy for classifying the various protocols can be found in [43] More recently, the delay bandwidth product problem arising out of high bandwidth networks and quality of service issues stemming from support of real time multimedia communication [7, 10, 12, 20, 21, 41] have added further complexities to the problem with QoS reigning as a unifying key theme. One of the lessons learned from congestion control research is that end to end rate based feedback control using various forms of linear increase exponential decrease can be effective, and asymmetry in the ....

Y. T. Wang and B. Sengupta. Performance analysis of a feedback congestion control policy under non-negligible propagation delay. In Proc. ACM SIGCOMM '91, pages 149--157, 1991.

....literature [BAE91, ECK91] especially in the context of B ISDN ATM networks, very few attempts have been made to quantify the effects of delayed information in feedback control, and even less to explicitly consider delays in the design of suitable well behaved feedback flow control mechanisms. In [WAN91] a threshold type feedback control policy is considered for a single queue with a deterministic server and finite buffering space, holding up to B packets. There are K users sending traffic to the queue. The round trip propagation delay for user k is k . The model is time slotted, each time slot ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy under Non-negligible Propagation Delay," in Proceedings of the ACM SIGCOM '91, pp. 149--157, Zurich, Switzerland, September 1991. References 59

....[1] etc. It is natural to model feedback flow control mechanisms by queueing systems with delayed feedback where the service times and job arrivals are time varying and statedependent. Unfortunately, the analysis of such systems is difficult, and only simple models have been considered (e.g. [12, 25, 2]) These difficulties have led to the development of approximate techniques. One such technique is the fluid or flow approximation, where the basic idea is to replace stochastic processes by corresponding deterministic processes which are related by means of differential or difference equations ....

Y-T. Wang, B. Sengupta, "Performance analysis of a feedback congestion control policy under non-negligible propagation delay", Proc. ACM SIGCOMM '91, Zurich, Switzerland,

.... Several studies have shown that adaptive rate control schemes are adversely affected by the presence of propagation delays between the controlled sources and the control nodes [18] and several authors have limited the applicability of such schemes to the realm of local area networks [19] [20] where propagation delays are no longer than several tens of the service time. Unlike in open loop schemes where no competition for network bandwidth takes place once a call is set up, an important characteristic of adaptive rate control schemes is that existing connections compete, at the cell ....

....difficulty, most previous analytical studies for adaptive rate control schemes have focused on aggregate performance measures. In addition to the large bandwidth propagation delay products characterizing high speed networks, an often overlooked by earlier discrete time studies such as [20], 14] characteristic is the potentially large disparity between the relative speeds of a network link and an input source. This disparity or speed up factor represents the time in slots required for a source to generate enough bits to form one cell 2 and is referred to as the source time ....

[Article contains additional citation context not shown here]

Y. T. Wang and B. Sengupta. Performance analysis of a feedback congestion control policy under non-negligible propagation delay. In Sigcomm'91, pages 149--157, Zurich, Switzerland, Sep 1991.

....This method is known as backpressure [Omi96] Yan95] The switch flow control mechanism is built into the input port of each switch. It uses a combination of XOFF XON thresholds and XOFF XON control frames to ensure that no data frames are dropped due to a lack of free memory in the switch [Wan91]. As will be expanded upon in Section 3, the flow control mechanism that has been proposed for use by switches, affects all sources and hence, as will be seen, this has an impact on the flow and congestion control procedure associated with the Transmission Control Protocol (TCP) running in each ....

....stored in input port(s) buffer. Hence if overload condition persist, the related input port buffers will eventually overflow [Hac96] As shown in Figure 2 the switch flow control compares the amount of data stored in the input port buffer (FIFO queue) with the XOFF threshold value (high water mark [Wan91]) If the XOFF threshold is reached, an XOFF control frame is sent to the adjacent switch (or the source) which in turn stops sending any new data frames. This causes the FIFO queue to release data until the current queue occupancy reaches the XON threshold (low water mark) At this point, an XON ....

Wang, Y. and Sengupta, B. (1991) Performance Analysis of a Feedback Congestion Control Policy under Non-Negligible Propagation Delay, Proceedings of ACM Sigcomm '91, 49-57.

....Since the amount of bandwidth mismatch depends on the propagation delay, increased propagation delay reduces the effectiveness of feedback based flow control schemes. The impact of increased propagation delay on the effectiveness of feedback based flow control schemes has been studied in the past [14], 15] 16] 17] In addition to the propagation delay, other system parameters may impact on the effectiveness of feedback based flow control schemes as well. In this paper, the impact of some system parameters namely, the network, VBR and ABR time scales will be investigated. t 1 t 2 t 1 ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under non-negligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

....period, the number of cells in transit becomes excessive and this causes cell losses. When an increase in the remaining bandwidth occurs, underutilization occurs for a duration of 2T d . While the effect of increased propagation delay on the flow control schemes has been studied in the past [15], 16] 17] almost no effort [18] has been focused on the study of the impact of other system parameters. In this paper, the impact of these system parameters (network, ABR and VBR time scales) on the feedback based flow control is investigated. The network time scale is defined as the ....

Y. T. Wang and B. Sengupta, "Performance Analysis of a Feedback Congestion Control Policy Under non-negligible Propagation Delay," ACM SIGCOMM'91, pp. 149-157, Sep. 1991.

No context found.

Y. T. Wang and B. Sengupta, Performance analysis of a feedback congestion control policy under nonnegligible propagation delay, Proceedings of ACM SIGCOMM, 149:157, (1991).

....at peak rate otherwise. To maximize throughput, we should try to have all feedback messages signal start as much as possible. Hence both the high and low thresholds should be as high as possible. In our simulation we choose the high threshold as B i Gamma D i for receiver i to avoid overflow [24], while the low threshold is chosen as B i Gamma D i Gamma Delta where Delta is used to avoid oscillation. B. Source Estimation (SE) Each receiver sends back both its queue length, q(t) and the queue growth rate, q(t) whenever q(t) changes sign; i.e. whenever it goes from increasing to ....

....queue length drops below its low threshold and all the other queues stay below their high thresholds. Otherwise it stops transmitting. In our simulation we choose the high threshold as B i Gamma D i for receiver i to avoid overflow and the low threshold as D i to try to avoid starvation [24]. We adopt a simple queue length estimation procedure taking advantage of the binary on off control and the fact that the source transmission peak rate is equal to the receiver throughput capability during the on period, i.e. Under this setting, the receiver queue size increases only if the ....

[Article contains additional citation context not shown here]

Y. T. Wang and B. Sengupta, "Performance analysis of a feedback congestion control policy under non-negligible propagation delay," in Proceedings of ACM SIGCOMM'91, pp. 149--157, 1991.

No context found.

Y. T. Wang and B. Sengupta. Performance analysis of a feed back congestion control policy. ACM SIGCOMM Computer Communication, pages 149--157, 1991.

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