D. P. Heyman, T. V. Lakshman, and L. Neidhardt, "A new method for analyzing feedback-based protocols with applications to engineering web traffic over the Internet, " in Proceedings of ACM SIGMETRICS'97, pp. 24--38, February 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....also considered by the model where the downlink is shared by the subscribers. Although we will not show the results due to space limitation, we have found that our analysis results in this paper can be applied to the reverse traffic model without any modifica tion. The similar model is treated in [9], but we consider RED and DRR as the packet scheduling algorithm in addition to the FIFO algorithm employed in [9] Further, we devote the fairness aspects of packet scheduling algorithms which are not consid ered in [9] This paper is organized as follows. In Section II, we describe the model ....

....due to space limitation, we have found that our analysis results in this paper can be applied to the reverse traffic model without any modifica tion. The similar model is treated in [9] but we consider RED and DRR as the packet scheduling algorithm in addition to the FIFO algorithm employed in [9]. Further, we devote the fairness aspects of packet scheduling algorithms which are not consid ered in [9] This paper is organized as follows. In Section II, we describe the model treated in Section III and IV. The packet scheduling algorithms is first summarized in Subsection II A. We will ....

[Article contains additional citation context not shown here]

L. Neidhardt D. P. Heyman, T. V. Lakshman, "A new method for analyzing feedback based protocols with applications to engineering web traffic over the internet," Proceedings oflEEE SIGMETRICS'97, pp. 2438, February 1997.

....random file sizes are independently and identically distributed. It can then be argued that the TCP controlled bandwidth sharing on the access link can be approximately modelled by using the Processor Sharing (PS) model from queueing theory (see [7] for an introduction to the PS model, and [10] [6], 5] 8] for studies involving the use of the PS model for TCP controlled bandwidth sharing) The PS model assumes fair bandwidth sharing; i.e. if there are n sessions active on the access link, then each session obtains exactly 1 n th of the bandwidth at all times. In practice, for TCP ....

....link, then each session obtains exactly 1 n th of the bandwidth at all times. In practice, for TCP controlled flows this is not true, especially if the file transfers are short, and if they encounter different propagation delays (for some studies of the applicability of such models to TCP, see [6] and [8] Let denote the rate of arrival of file transfer requests, let X denote the random file size (EX denotes the expected file size) and let b denote the Internet access link bandwidth. For the purpose of analysis we assume a fluid model. For this model we can obtain the time average ....

D.P. Heyman, T.V. Lakshman, A.L. Neidhardt, "A New Method for Analyzing Feedback-Based Protocols with Applications to Engineering Web Traffic over the Internet," Performance Evaluation Review, Vol. 25, No. 1, 1997.

....the error becoming larger as ff approaches 1. We trace these observations to the fact that TCP performance is sensitive to packet loss. We find that with heavy tailed file sizes, there is greater buffer loss, and hence throughput suffers. The most important related work is that of Heyman et al. [9]. The authors study a model comprising several hosts connected to a backbone link via low speed access links. Each host alternates between downloading files and thinking . An analytical model is developed and it is shown that the session throughputs are insensitive to the file size distributions. ....

....and it is shown that the session throughputs are insensitive to the file size distributions. There is, however, no analytical study of LRD in the network traffic. Another related reference is [3] where the authors provide asymptotic approximations for bandwidth dimensioning using the model in [9]. Our work demonstrates the importance of considering heavy tailed file sizes in network buffer sizing for TCP controlled elastic traffic, and relates the parameter of the Pareto file size distribution to the Hurst parameter of the traffic in the network link. The paper is organised as follows. ....

[Article contains additional citation context not shown here]

D.P. Heyman, T.V. Lakshman, A.L. Neidhardt, "A New Method for Analyzing Feedback-Based Protocols with Applications to Engineering Web Traffic over the Internet, " Performance Evaluation Review, Vol. 25, No. 1, Proc. ACM SIGMETRICS '97, pp 24-38, 1997.

....and networks, see for example [1] or [2] and our model is one of the simplest; though the entity modeled by a job is non standard, see Section 2. Our CQN model consists of one inifinite server (IS) and one processor sharing (PS) server and was motivated by Heyman, Lakshman, and Neidhardt, [3], who also use a CQN model where the focus is on analyzing the performance of the closedloop control TCP in the context of Web traffic. An important practical feature of these CQN models is their insensitivity: the distribution of the underlying random variables influences the performance only via ....

.... UBR or ABR: Significant degradation in throughput has been observed and various enhancements have been considered, see for example Romanov and Floyd [19] and Fang et al. 20] The previously cited Heyman Lakshman Neidhardt s model predicts behavior both when the performance is poor and good, [3]. Balakrishnan, Rahul Seshan propose an integrated congestion management architecture that provides feedback to the application layer and applies to UDP as well as TCP, 21] Thus, we view as complementary: 1) network design that assumes well performing closed loop controls, and (2) control ....

[Article contains additional citation context not shown here]

D. P. Heyman, T. V. Lakshman, & A. L. Neidhardt, "A new method for analyzing feedback-based protocols with applications to engineering web traffic over the internet," Performance Evaluation Review, Vol. 25, no. 1, Proc. ACM-SIGMETRICS'97, 1997, pp. 24-38.

....dynamics of TCP instead of details of its protocol beh viors ths directly impactth QoS perceived by end users. A more appropriate model for IP network design th ereforesh ould focus on transactionlevel issues such asth e#ect of arrival departure dynamics on long term th ough## of TCP[3] Recently, [4] and [5]h ave studiedth problem of link dimensioning for a single bottleneck link s h red by a fi ed number of ON OFF type of tra#c sources. During an ON period, a tra#c source transmits a random amount of data using TCP. By a separation of time scale argument,th2 assume th tth link bandwidth is ....

D. Heyman, T. V. Laksh# n and A. Niedh rt. "A new meth od for analyzing feedback-based protocols with applications to engineering web tra#c over th Internet." In Sigmetrics 97, page 24-38, 1997.

....condition is known and stable, we should try to avoid slow start, and do congestion avoidance directly. This can be achieved by setting the initial ssthresh to be no larger than the initial cwnd. To find cwnd opt , we only need to consider the congestion avoidance phase. Figure 2 (obtained from [12]) depicts roughly how the congestion window and service rate evolve during congestion for TCP Reno, going through periodical epochs. The sender has a loss at the end of each epoch, and recovers from it using fast retransmission. Window Size Time Service Rate Figure 2: Evolution of service rate ....

D. P. Heyman, T.V. Lakshman, and A. L. Neidhardt. A New Method for Analyzing Feedback-Based Protocols with Applications to Engineering Web Tra#c over the Internet. In Proc. ACM SIGMETRICS '97, 1997. 20

....cwnd of b Wc 1 2 c and has an infinite amount of data to send. Define S(t) as the service rate it receives at time t, and W (t) as its congestion window size at t. Figure 1 depicts roughly how S(t) and W (t) evolve with time, going through periodical epochs. A similar figure also appeared in [15]. For simplicity, we ignore the time required by the fast retransmission algorithm to wait for 3 duplicated ACK s. Let E: the duration of an epoch FE : the number of packets sent during an epoch b: the average number of packets acknowledged by an ACK Close inspection of the congestion ....

D. Heyman, T. Lakshman, and A. Neidhardt, "A New Method for Analyzing Feedback-Based Protocols with Applications to Engineering Web Traffic over the Internet," Proc. SIGMETRICS '97, 1997.

....is limited by the access link capacities. However, when the number of active sources is greater than C; then we assume that there is some congestion control mechanism that attempts to divide the capacity fairly among the competing sources. This model was proposed and analyzed for Web traffic in [16]. 2 Bottleneck Node C Source Source Source Source Access link, line rate=1 Figure 1: Instantaneous Model of Congestion Controlled Best Effort Traffic The interest in such a model is due to the dramatic development of best effort services over the last few years. Unlike traditional telephone ....

....networks, the available bit rate (ABR) service performs the congestion control function, again through a feedback control mechanism [4, 30, 19, 1, 2, 5] In either case, typically the sharing mechanism that allocates the node s capacity to the competing sources is not efficient. For instance, [16] presents an analytical model to quantify this inefficiency in the case of TCP. According to their model, when the number of active sources is greater than C; a fraction (denoted by fi) of the total capacity C is shared equally by the active sources. Thus, the factor fi will depend on the ....

[Article contains additional citation context not shown here]

D. Heyman, T. V. Lakshman, and A. Niedhart. A new method for analyzing feedback-based protocols with applications to engineering web traffic over the internet. In Sigmetrics 97, pages 24--38, 1997.

....control mechanism that attempts to divide the capacity fairly among Bottleneck Node C Source Source Source Source Access link, line rate=1 Figure 1: Instantaneous Model of Congestion Controlled Best Effort Traffic the competing sources. This model was proposed and analyzed for Web traffic in [16]. The interest in such a model is due to the dramatic development of best effort services over the last few years. Unlike traditional telephone networks, best effort traffic is not allocated a fixed bandwidth in the network. Rather, the available capacity is divided among various competing sources ....

....the available bit rate (ABR) service performs the congestion control function, again through a feedback control mechanism [3, 25, 19, 1, 2, 4] In either case, typically the sharing mechanism that allocates the bottleneck node s capacity to the competing sources is not efficient. For instance, [16] presents an analytical model to quantify this inefficiency in the case of TCP. We do not pursue this here. We simply assume that, when the number of active sources is greater than C; a fraction (denoted by fi) of the total capacity C is shared equally by the active sources. Thus, the factor fi ....

[Article contains additional citation context not shown here]

D. Heyman, T. V. Lakshman, and A. Niedhart. A new method for analyzing feedback-based protocols with applications to engineering web traffic over the internet. In Sigmetrics 97, pages 24--38, 1997.

....performing control is that it maintains some bytes in queue at the bottleneck link, with minimal packet or cell loss. Although current control implementations do not necessarily meet this assumption, ongoing research efforts will lead to improved controls, see for example Heyman LakshmanNeidhardt [3] and Floyd [4] We view as complementary: 1) network design that assumes well performing closed loop controls, and (2) control implementations that make good use of the deployed bandwidth. Given well performing controls and for periods of high load, which is the case of interest for engineering ....

....present. Gamma1 k = f k =B, where f k is the mean file size of a class k connection and B is the engineered bandwidth. Thus, a class k connection is characterized by k and f k . Moreover, for the distribution of jobs at the PS node only the product u k = k f k is pertinent. Heyman et al. [3] in their modified Engset models also find that the impact on performance by the distribution of file sizes and think times is only via the product u k . u k , in bits per second, represents the throughput averaged over both active and idle periods, for a class k source, given that the target link ....

D. P. Heyman, T. V. Lakshman, & A. L. Neidhardt, A new method for analyzing feedback-based protocols with applications to engineering web traffic over the internet, Performance Evaluation Review, Vol. 25, no. 1, Proc. ACM-SIGMETRICS'97, 1997, pp. 24-38.

....and networks, see for example [3] or [4] and our model is one of the simplest; though the entity modeled by a job is non standard, see Section 2. Our CQN model consists of one inifinite server (IS) and one processor sharing (PS) server and was motivated by Heyman, Lakshman, and Neidhardt, [5], who also use a CQN model where focus is on analyzing the performance of the closed loop control TCP in the context of Web traffic. An important practical feature of these CQN models is their insensitivity: the distribution of the underlying random variables influences the performance only via ....

.... UBR or ABR: Significant degradation in throughput has been observed and various enhancements have been considered, see for example Romanov and Floyd [19] and Fang et al. 20] The previously cited Heyman Lakshman Neidhardt s model predicts behavior both when the performance is poor and good, [5]. Thus, we view as complementary: 1) network design that assumes well performing closed loop controls, and (2) control implementations that make good use of the deployed bandwidth. 1.2 Outline of Remaining Sections Section 2 describes the CQN model and performance objectives. Section 3 ....

[Article contains additional citation context not shown here]

D. P. Heyman, T. V. Lakshman, & A. L. Neidhardt, "A new method for analyzing feedback-based protocols with applications to engineering web traffic over the internet," Performance Evaluation Review, Vol. 25, no. 1, Proc. ACM-SIGMETRICS'97, 1997, pp. 24-38.

No context found.

D. P. Heyman, T. V. Lakshman, and L. Neidhardt, "A new method for analyzing feedback-based protocols with applications to engineering web traffic over the Internet, " in Proceedings of ACM SIGMETRICS'97, pp. 24--38, February 1997.

No context found.

D. P. Heyman, T. V. Lakshman, and L. Neidhardt, "A new method for analyzing feedback-based protocols with applications to engineering web traffic over the Internet," in Proceedings of ACM SIGMETRICS '97, pp. 24--38, Feb. 1997.

No context found.

L. N. D. P. Heyman, T. V. Lakshman, "A new method for analyzing feedback--based protocols with applications to engineering web traffic over the internet," Proceedings of IEEE SIGMETRICS'97, pp. 24--38, February 1997.

No context found.

D. P. Heyman, T. V. Lakshman, & A. L. Neidhardt, A new method for analyzing feedback-basedprotocols with applications to engineering web tra#c over the internet, Performance Evaluation Review, Vol. 25, no. 1, Proc. ACM-SIGMETRICS'97, 1997, pp. 24-38.

No context found.

Heyman, D. P., Lakshman, T. V., and Neidhardt, A. L. A new method for analyzing feedback-based protocols with applications to engineering web traf- c over the Internet. In Proc. ACM SIGMETRICS (1997).

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