T. Tuan, and K. Park, "Multiple Time Scale Congestion Control for Self-similar Network Tra#c", Performance Evaluation 36-37(1-4), pp. 359-386, 1999.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....side, the MAI self similarity implies that there exists a nontrivial predictive MAI structure at coarse time scales, which can be exploited for interference management to improve system performance. Self similarity in network traffic has been utilized for congestion control in wireline networks [20]. Exploiting the predictive MAI structure, we develop a rate adaptation scheme to ameliorate the system performance. Our rate control can be summarized as follows: If the (predicted) future interference is weak, we increase the transmission rate via decreasing the spreading gain or increasing ....

....MAI may not hold; on the other hand, if Tm is too large, it is difficult to imagine a mechanism that would be able to effectively exploit the prediction, simply because during such a long period, too many changes could occur, yielding zero net gain for control decisions. Following the lines of [20], we use the off line entropy method to determine Tm . Given a time scale Tm 0, we define V 1 = i#[t Tm ,t] X i , V 2 = i#[t,t Tm ] X i . 13) We introduce two random variables L 1 , L 2 as the quantization of V 1 , V 2 , i.e. L j = L j (V j ) L j [1, M ] j = 1, 2, 14) where M ....

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," Performance Evaluation, vol. 36, pp. 359--386, 1999.

....is of great importance for various network applications. Knowledge of the available bandwidth on an end to end path can improve rate based streaming applications [1] end to end admission control [2] server selection [3] optimal route selection in overlay networks [4] congestion control [5] as well as service level agreement verification [6] Obtaining useful estimates of the available bandwidth from routers is often not possible due to various technical and privacy issues or due to an insufficient level of measurement resolution or accuracy. Thus, it becomes necessary to infer the ....

T. Tuan and K. Park, "Multiple time scale congestion control for self-similar network traffic," in Performance Evaluation, vol. 36, pp. 359--386, 1999.

....is of great importance for various network applications. Knowledge of the available bandwidth on an end to end path can improve rate based streaming applications [1] end to end admission control [2] server selection [3] optimal route selection in overlay networks [4] congestion control [5] as well as service level agreement verification [6] Obtaining useful estimates of the available bandwidth from routers is often not possible due to various technical and privacy issues or due to an insufficient level of measurement resolution or accuracy. Thus, it becomes necessary to infer the ....

T. Tuan and K. Park, "Multiple time scale congestion control for self-similar network traffic," in Performance Evaluation, vol. 36, pp. 359--386, 1999.

....Most research into control mechanisms for elastic traffic has ignored the long range dependence of network traffic [4] 8] and the transmission rates were determined without reference to the correlation structure of the traffic. Tuan and Park s multiple time scale congestion control algorithm [9] is a notable exception. Their proposal modifies the well known linear increase multiplicative decrease concept used in the TCP algorithm so that long term trends in network traffic are accounted for. However, they did not explicitly use the correlation structure of the long range dependent ....

T. Tuan and K. Park, "Multiple time scale congestion control for self-similar network traffic," Perform. Eval., vol. 36--37, no. 1--4, pp. 359--386, 1999.

....shared losses. Their approach is more effective with active probing, rather than in band measurements. To avoid problems with collecting delay information and clock synchronization, correlation among TCP round trip time (RTT) estimates (e.g. 6] or throughput estimates at the sender (e.g. [23]) may substitute one way delays. Although using these metrics eliminates any changes (e.g. timestamping support) to the receiver, the delay on the forward path cannot be isolated from that of the reverse path and the delays at the receivers themselves, as discussed in the next section. 3 ....

T. Tuan and K. Park. Multiple Time Scale Congestion Control for Self-Similar Network Traffic. Performance Evaluation, 36:359--386, 1999.

....side, the MAI self similarity implies that there exists a nontrivial predictive MAI structure at coarse time scales, which can be exploited for interference management to improve system performance. Self similarity in network traffic has been utilized for congestion control in wireline networks [19]. Exploiting the predictive MAI structure, we develop a rate adaptation scheme to ameliorate the system performance. Our rate control can be summarized as follows: If the (predicted) future interference is weak, we increase the transmission rate via decreasing the spreading gain or increasing ....

....MAI may not hold; on the other hand, if T# is too large, it is difficult to imagine a mechanism that would be able to effectively exploit the prediction, simply because during such a long period, too many changes could occur, yielding zero net gain for control decisions. Following the lines of [19], we use the off line entropy method to determine T# . Given a time scale T# #,we define V # # # ####### ### I#i#; V # # # ######### # I#i#: 13) We introduce two random variables L # ;L # as the quantization of V # ;V # ,i.e. L # # L # #V # #; L # # ##;M#; j ##; #; 14) where M denotes ....

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," Performance Evaluation, vol. 36, pp. 359--386, Aug. 1999.

....Advanced Research Projects Agency or the U. S. Government. volves binary feedback [7] and proportional controllers [8] for ATM (Asynchronous Transfer Mode) networks and linearincrease exponential decrease controllers for TCP IP (Transmission Control Protocol Internet Protocol) networks (see [15] for a recent version) Recent rate based approaches attempt to achieve better performance by incorporating control theoretic techniques, including proportional derivative (PD) controllers [9] 10] 17] 30] and those using optimal control and dynamic game techniques such as linear quadratic ....

....an auto regressive moving average process corrupted by a sequence of independent and identically distributed random numbers with zero mean and finite variance. Compared with [12] 13] MMF models have more structure, and better performance is therefore expected when such models are available. In [15], the authors incorporate a long range dependent model into the design of a linear increase exponential decrease controller. Our MMF model yields to a decision theoretic analysis, as mentioned above, resulting in a controller that is not constrained to be linear increase exponential decrease. ....

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," to appear in Performance Evaluation, 1999.

....credits to those sources. Early rate based work involves binary feedback [9] and proportional (P) controllers [10] 11] for ATM (Asynchronous Transfer Mode) networks and linear increase exponential decrease controllers for TCP IP (Transmission Control Protocol Internet Protocol) networks (see [18] for a recent version) More recent rate based approaches attempt to achieve better performance by incorporating control theoretic techniques, including proportional derivative (PD) controllers [12] 13] 20] 33] and those using optimal control and dynamic game techniques such as linear ....

....by an auto regressive moving average process corrupted by a sequence of independent and identically distributed random numbers with zero mean and nite variance. Compared with [15] 16] MMF models have more structure and better performance is therefore expected when such models are available. In [18], the authors incorporate a long range dependent model into the design of a linear increase exponential decrease controller. Our MMF model yields to a decision theoretic analysis, as mentioned above, resulting in a controller that is not constrained to be linear increase exponential decrease. ....

T. Tuan and K. Park, \Multiple time scale congestion control for self-similar network trac," to appear in Performance Evaluation, 1999.

....this would require maintaining overwhelming amounts of per ow state information. It becomes necessary to infer the properties from edge based measurements, which are relatively easy and inexpensive to make. In this light, several authors have proposed edge based techniques for congestion control [1 3], estimating the bottleneck bandwidth [4 6] inferring multicast This work was supported by the NSF, grant no. ANI 9979465, by ONR, grant no. N00014 99 10813, by DARPA, grant no. R 36400, and by Texas Instruments. routing trees [6] performing admission control [7] and detecting ows with the ....

....bandwidth [9,10] in that it is model based. By employing short bursts of packets these techniques are naturally restricted to estimating cross trac only over short time periods [9] Other methods capable of dealign with larger time scales use only an indirect measure of the competing trac load [3]. Delphi instead uses probes spaced farther apart and improves the accuracy in cross trac estimates by leveraging statistical knowledge of network dynamics provided by a versatile trac model, the multifractal wavelet model (MWM) 11] Aggregated trac on a link has been shown to be multiscale in ....

T. Tuan and K. Park, \Multiple time scale congestion control for self-similar network trac," in Performance evaluation, vol. 36, pp. 359-386, 1999.

....control for best effort traffic. Earlier work involves binary feedback [9] and proportional (P) controllers [10] 11] for ATM (Asynchronous Transfer Mode) networks and linear increase and exponentialdecrease type of controllers for TCP IP (Transmission Control Protocol Internet Protocol) networks [18]. More recently, to achieve better performance several control theoretic approaches have been studied, including proportional derivative (PD) controllers and their variants [12] 13] 20] and those using optimal control and dynamic game techniques such as linear quadratic (LQ) team, H 1 , and ....

....at the bottleneck node by an auto regressive moving average process corrupted by a sequence of independent and identically distributed random numbers with zero mean and finite variance. Compared with [15] 16] our MMF model has more structure and better performance is therefore expected. In [18], the authors incorporate a long range dependent model into the design of a linear increase exponential decrease type controller. Our MMF model yields to a decision theoretic analysis, as mentioned above, resulting in controller that is not constrained to be of linear increase exponential decrease ....

T. Tuan and K. Park, "Multiple time scale congestion control for self-similar network traffic," to appear in Performance Evaluation, 1999.

No context found.

T. Tuan and K. Park, "Multiple time scale congestion control for self-similar network tra#c," Perf. Eval., vol. 36-37, pp. 359--386, 1999.

No context found.

TUAN,T.AND PARK, K. 1999. Multiple time scale congestion control for self-similar network traffic. Perf. Eval. 36, 359--386.

.... Long range dependence and self similarity of aggregate trac can be shown to persist at multiplexing points in the network as long as connection durations or object sizes being transported are heavy tailed, insensitive to details in the protocol stack or network con guration [14] In previous work [19], we explored the feasibility of exploiting long range correlation structure in self similar network trac for congestion control. We introduced the framework of Multiple Time Scale Congestion Control (MTSC) and showed its e ectiveness at enhancing performance for rate based feedback control. In ....

....to changes in large time scale network state, we choose a time scale closer to 1 second than 10 second. We use a 2 second time scale for this reason in the rest of the paper. 3. Multiple Time Scale TCP 3.1. Multiple Time Scale Congestion Control We use Selective Slope Control (SSC) [19] to engage large time scale correlation structure when modulating the trac control behavior of a feedback congestion control. SSC adjusts the slope of linear increase during the linear increase phase of AIMD congestion controls based on the predicted large time scale network state. If network ....

[Article contains additional citation context not shown here]

T. Tuan and K. Park. Multiple time scale congestion control for self-similar network trac. Performance Evaluation, 36:359-386, 1999.

....of buffer capacity and details in the protocol stack or network configuration [Feldmann et al. 1998; Park et al. 1996] How to effectively utilize large time scale, probabilistic information afforded by traffic characteristics to improve performance is a nontrivial problem. In previous work [Tuan and Park 1999], we have explored the feasibility of exploiting long range correlation structure in self similar network traffic for congestion control. We introduced the framework of Multiple Time Scale Congestion Control (MTSC) and showed its effectiveness at enhancing performance for ratebased feedback ....

....in large time scale network state, we choose a time scale closer to 1 second than 10 second. We use a 2 second time scale for this reason in the rest of the paper. 3. MULTIPLE TIME SCALE TCP 3. 1 Multiple Time Scale Congestion Control The framework of multiple time scale congestion control [Tuan and Park 1999], in general, allows for n level time scale congestion control for n 1 where information extracted at n separate time scales is cooperatively engaged to modulate the output behavior of the feedback congestion control residing at the lowest time scale (i.e. n = 1) The ultimate goal of MTSC is to ....

[Article contains additional citation context not shown here]

Tuan, T. and Park, K. 1999. Multiple time scale congestion control for self-similar network traffic. Performance Evaluation 36, 359--386.

....cantly di erent from those that stem from assuming independence. This has been con rmed and further analyzed by Cidon et al. 15,16] albeit excluding LRD input processes. In most recent work [57] we have extended adaptive redundancy control in the framework of multiple time scale trac control [56] to exploit large time scale predictability structure of self similar trac which imparts additional proactivity. An application to TCP congestion control can be found in [43] 3 Network Model 3.1 Packet level FEC The goal of adaptive forward error correction is to adjust the level of redundancy ....

....are no instrinsic reasons to suspect that AFEC would not become a well behaved member of the existing protocol family. A gametheoretic discussion of QoS provision issues can be found in [13,14] In most recent work [57] we have extended AFEC to the multiple time scale trac control framework [56] a novel workload sensitive trac control paradigm where, in addition to AFEC s adaptive feedback control, long range correlation structure in self similar trac is explicitly exploited to a ect further performance improvement on top of AFEC. ....

T. Tuan and K. Park. Multiple time scale congestion control for self-similar network trac. Performance Evaluation, 36:359-386, 1999.

....(e.g. TCP) This is relevant in broadband wide area networks where the delay bandwidth product problem is especially severe, and mitigating the performance degradation due to outdated feedback critical to facilitating scalable, adaptive traffic control. The initial success of this approach [67, 68, 62, 69] (see Chapter 18 for an application to rate based congestion control) leads to a generalization to workload sensitive traffic control where facilitation of workload sensitivity is expanded along several traffic control dimensions including the two core features for harnessing predictability at ....

....the time scale of the feedback loop i.e. round trip time is an order of magnitude (or more) smaller than the time scale at which longrange correlation structure, in practice, manifests itself: millisecond vs. second range. The multiple time scale traffic control framework was introduced in [67], and shown to be effective at yielding significant performance improvement when large time scale correlation is exploited for traffic control. In [67] see also Chapter 18) large time scale correlation structure was on line estimated and utilized to modulate the bandwidth consumption behavior of ....

[Article contains additional citation context not shown here]

T. Tuan and K. Park. Multiple time scale congestion control for self-similar network traffic. Performance Evaluation, 36:359--386, 1999.

.... a vis bandwidth for self similar traffic, and the consequent role of short range correlations in affecting first order performance characteristics when buffer capacity is indeed provisioned to be small [29, 58] On the feedback control side is the work on multiple time scale congestion control [67, 68] which tries to exploit correlation structure that exists across multiple time scales in self similar traffic for congestion control purposes. In spite of the negative performance impact of self similarity, on the positive side, long range dependence admits the possibility of utilizing correlation ....

....is a nontrivial technical challenge for two principal reasons: one, the correlation structure in question exists at time scales typically an order of magnitude or more above that of the feedback loop, and two, the information extracted is necessarily imprecise due to its probabilistic nature 4 . [67, 68] show that large time scale correlation structure can be employed to yield significant performance gains both for throughput maximization using TCP and rate based control and end to end QoS control within the framework of adaptive redundancy control [52, 68] An important by product of this ....

[Article contains additional citation context not shown here]

T. Tuan and K. Park. Multiple time scale congestion control for self-similar network traffic. Performance Evaluation, 36:359--386, 1999.

....are heavy tailed, irrespective of buffer capacity and details in the protocol stack or network configuration [14, 26] How to effectively utilize large time scale, probabilistic information afforded by traffic characteristics to improve performance is a nontrivial problem. In previous work [37], we have explored the feasibility of exploiting long range correlation structure in self similar network traffic for congestion control. We introduced the framework of Multiple Time Scale Congestion Control (MTSC) and showed its effectiveness at enhancing performance for rate based feedback ....

....in large time scale network state, we choose a time scale closer to 1 second than 10 second. We use a 2 second time scale for this reason in the rest of the paper. 3 Multiple Time Scale TCP 3. 1 Multiple Time Scale Congestion Control The framework of multiple time scale congestion control [37], in general, allows for n level time scale congestion control for n 1 where information extracted at n separate time scales is cooperatively engaged to modulate the output behavior of the feedback congestion control residing at the lowest time scale (i.e. n = 1) The ultimate goal of MTSC is to ....

[Article contains additional citation context not shown here]

T. Tuan and K. Park. Multiple time scale congestion control for self-similar network traffic. Performance Evaluation, 36:359--386, 1999.

....In this paper we show that the aforementioned problems self similar burstiness and feedback redundancy control with long RTTs can be effectively addressed yielding significant performance improvements. Our solution is based on the framework of Multiple Time Scale Congestion Control (MTSC) [29], 30] which has been recently advanced in the context of throughput maximization. The basic premise of MTSC hinges on the fact that despite the detrimental performance effect associated with self similar burstiness the presence of nontrivial correlation structure across multiple time scales ....

....scale congestion control where the large time scale module CL separated by an order of magnitude from the small time scale module C S is coupled to the latter to yield a new control CL Omega S . For throughput maximization, for example, the coupling takes on a multiplicative form [29]. For QoS control us Low Contention l L High Contention l H High DC Level Low DC Level Level Shift Fig. III.2. Additive coupling via selective DC level adjustment i.e. level shift between high and low contention periods. ing adaptive FEC, we employ additive coupling. The latter is ....

[Article contains additional citation context not shown here]

T. Tuan and K. Park. Multiple time scale congestion control for selfsimilar network traffic. Performance Evaluation, 36:359--386, 1999.

No context found.

T. Tuan, and K. Park, "Multiple Time Scale Congestion Control for Self-similar Network Tra#c", Performance Evaluation 36-37(1-4), pp. 359-386, 1999.

No context found.

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," to appear in Performance Evaluation, 1999.

No context found.

T. Tuan and K. Park. Multiple Time Scale Congestion Control for Self-Similar Network Traffic. Performance Evaluation, 36:359--386, 1999.

No context found.

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," to appear in Performance Evaluation, 1999.

No context found.

T. Tuan and K. Park, "Multiple Time Scale Congestion Control for SelfSimilar Network Traffic," Performance Evaluation, vol. 36, pp. 359--386, 1999.

No context found.

Tsunyi Tuan, Kihong Park, Multiple Time Scale Congestion Control for Self-Similar Network Tra#c, Network Systems Lab, Department ofComputer Sciences, Purdue University, West Lafayette, IN4790H USA, Preprint submitted to Elsevier Preprint.

No context found.

T. Tuan and K. Park, "Multiple time scale congestion control for selfsimilar network traffic," Performance Evaluation, vol. 36-37, pp. 359-- 386, 1999.

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