X. Zhang and K. G. Shin, Integrated Rate and Credit Feedback Control for ABR Service in ATM Networks, Technical Report, EECS, Univ. of Michigan, July 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....at the source and a queue length process Q(t) at the bottleneck node, respectively. Due to its simplicity, effectiveness, and approximation accuracy (particularly for heavy traffic) the fluid modeling has been used effectively for the analysis and evaluation of Bg unicast flow control schemes [9 17]. The existence of multiple paths in an B t pt to mpt connection complicates its modeling and analysis. As in any feedback control system, the gM cell round trip delay plays a critical role in determining system performance. In all previous analyses of unicast flow control using the fluid model, ....

....system, the gM cell round trip delay plays a critical role in determining system performance. In all previous analyses of unicast flow control using the fluid model, the round trip delay is treated as a constant equal to the value found determined during the setup of each s t connection [9 17]. However, as mentioned earlier, the m cell round trip delay in a pt to mpt connection varies significantly with time. Our model, therefore, takes this variation into account. By applying the proposed second order rate control (to be discussed in Section 4) the source rate control can be adapted ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM, April 1997. 25

....to implement as compared to the CI based scheme. In this paper, we will focus only on the CI based scheme modeling and analysis (and report the results on ER based scheme in another paper) We model the CI based flow control system by the first order fluid approximation method [8] 9] 10] [11], which uses the continuous time functions R(t) and Q(t) as the fluid approximations of the source rate and bottleneck queue length, respectively. We also assume the existence of only a single bottleneck 2 at a time with queue length equal to Q(t) and a persistent source (the most stressful ....

....the current time and t 0 is the time of the last rate update; Q h (Q l ) is the high (low) queue threshold for the multicast tree bottleneck s buffer, and = minf 1 ; 2 ; Delta Delta Delta ; ng is the multicast tree bottleneck bandwidth. IV. THE SECOND ORDER RATE CONTROL As discussed in [11], increasing or decreasing R(t) is not effective enough to have the maximum queue length Qmax upper bounded by the maximum buffer capacity Cmax when the multicast tree RM cell RTT varies due to drift of multicast tree bottleneck. This is because rate increase decrease control can only make R(t) ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM,

....at the source and a queue length process Q(t) at the bottleneck node, respectively. Due to its simplicity, effectiveness, and approximation accuracy (particularly for heavy traffic) the fluid modeling has been used effectively for the analysis and evaluation of abr unicast flow control schemes [9 17]. The existence of multiple paths in an abr pt to mpt connection complicates its modeling and analysis. As in any feedback control system, the rm cell round trip delay plays a critical role in determining system performance. In all previous analyses of unicast flow control using the fluid model, ....

....system, the rm cell round trip delay plays a critical role in determining system performance. In all previous analyses of unicast flow control using the fluid model, the round trip delay is treated as a constant equal to the value found determined during the setup of each abr connection [9 17]. However, as mentioned earlier, the rm cell round trip delay in a pt to mpt connection varies significantly with time. Our model, therefore, takes this variation into account. By applying the proposed second order rate control (to be discussed in Section 4) the source rate control can be adapted ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM, April 1997.

.... only on the CI based scheme modeling and analysis (and report the results on ER based scheme in another paper) We model the CI based scheme by using the firstorder fluid approximation, which characterizes the flow control system with coupled time delayed differential equations [8] 9] 10] [11]. We also assume the existence of only a single bottleneck 2 at a time with queue length equal to Q(t) and a persistent source with ACR = R(t) for each multicast connection. Such a data source model enables us to examine the proposed scheme under the most stressful condition. A. System ....

....the current time and t 0 is the time of the last rate update; Q h (Q l ) is the high (low) queue threshold for the multicast tree bottleneck s buffer, and = minf 1 ; 2 ; Delta Delta Delta ; ng is the multicast tree bottleneck bandwidth. IV. THE SECOND ORDER RATE CONTROL As discussed in [11], increasing or decreasing R(t) is not effective enough to have Qmax upper bounded by the maximum buffer capacity Cmax when the multicast tree RM cell RTT varies due to drift of multicast tree bottleneck. This is because rate increase decrease control can only make R(t) fluctuate around the ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks", in Proc. of IEEE INFOCOM,

....average of the offered traffic load does not grow beyond the bottleneck bandwidth, we set ,5 (1 ) 10] throughout the rest of the paper. In the following we present some of the numerical results we obtained to evaluate system performance and the more complete and detailed results can be found in [12]. Performance Analysis for Pattern I, II, and III: As expected, Figure 4 shows Q . increases monotonically with c and . 2 . also increases roughly linearly with c and Q . increases faster for a larger . In Figure 5, R is found to decrease monotonically as c and increase, and to ....

....will focus on these two patterns. Notice that for these two patterns, when Q . t) restarts rate increase from with a smaller increase rate of c x instead of c. The detailed descriptions of control patterns and derivations of their corresponding analytical expressions are available in [12]. Here we only present some numeri cal results on the transient state performance. The net work condition remains the same as in Section 5.2. But we use 0 = 4p and = log 2 here. In Figure 8 9, we observe that for a given a larger c not only results in a higher transient state average ....

X. Zhang and K. G. Shin, Integrated Rate and Credit Feedback Control for ABR Service in ATM Networks, Technical Report, EECS, Univ. of Michigan, July 1996.

....product is large. However, the ER based scheme s superiority to the CI based scheme comes at the expense of complexity. We model the proposed CI based scheme by using the first order fluid approximation, which characterizes the flow control system with coupled time delayed differential equations [8 14]. We assume the existence of only a single bottleneck 2 at a time with queue length Q(t) and a persistent source with ACR R(t) for each multicast connection. Such a data source model enables us to study the proposed scheme under the most stressful condition. Figure 5 depicts the system model ....

....1 t is then given by lJmt o 7 f R(v)dv. 3. 3 Analytic Solutions of Multicast tree Bottleneck Path Dynamics Here we only present the analytical expressions of the key performance measures for the dynamics of multicast tree bottleneck path, and omit their derivations, which can be found in [14]. 1) Maximum Queue Length of the multicast tree bottleneck path: Q. at dt (R.e O )k p)dt = a T A A P (3.3) J0 2 where T,a :r . Qa Td a log (1 T,ax) and Ra aTa. v 0 ) 2) Oscillation Period of the multicast tree bottleneck path: A log = where T is non negative real ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of [EEE [NFOCOM, April 1997.

....otherwise (if (t,t 0 ) # #) by letting (t,t 0 ) be small enough such that multicast tree bottleneck path that the traffic source can perceive is unique during (t , t 0 ) shown in [7] # s swing between # min and # max is still large enough to make a significant impact on Qmax . As discussed in [15], increasing or decreasing R#t# is not effective enough to have the maximum queue length Qmax upper bounded by the maximum buffer capacity Cmax when the multicast tree RMcell RTT # varies due to drift of the multicast tree bottleneck. This is because rate increase decrease control can only make ....

.... This is because rate increase decrease control can only make R#t# fluctuate around the designated bandwidth, but cannot adjust the rate fluctuation amplitude that determines Qmax.So, Qmax also depends on the source rate gain parameter # (to be detailed in Section V) Qmax is analytically shown in [15] to increase with both # and rate gain parameter # = dR#t# dt and can be written as a function, Qmax ##; # #,orQmax ### for a given # . In reality, the buffer capacity, Cmax , on the bottleneck path is finite, and hence, to ensure cell lossless transmission, the condition Qmax # Cmax must ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM, pp. 1297--1305, April 1997.

....two decoupled components: flow control and error control mechanisms. A. The Flow Control Mechanism The flow control is to dynamically adapt user demand to currently available network resources. The network resources consists of two parts: bandwidth capacity and buffer capacity. As discussed in [10], the traditional AIMD rate control, which only applies direct increasing decreasing control (first order rate control) over source rate R(t) is not effective enough to have the maximum queue length Q max upper bounded by the maximum buffer capacity C max . This is because the first order rate ....

....the first order rate control can only make R(t) fluctuate around the designated bandwidth, but cannot adjust the rate fluctuation amplitude that determines Q max . Consequently, the first order rate control only exercises the control over bandwidth, but leaves bottleneck buffers uncontrolled. In [10] and [5] Q max is analytically shown to increase with both the rate gain parameter and the connection s RTD. In [5] we developed the second order rate control, called ff control, to deal with the RTD variation due to the bottleneck drift in a multicast communication tree. In this paper, we ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM, pp. 1297--1305, April 1997. Full paper version: Tech. Report, CSE-TR-354-97, The University of Michigan.

....two decoupled components: flow control and error control mechanisms. A. The Flow Control Mechanism The flow control is to dynamically adapt user demand to currently available network resources. The network resources consists of two parts: bandwidth capacity and buffer capacity. As discussed in [10], the traditional AIMD rate control, which only applies direct increasing decreasing control (first order rate control) over source rate R#t#, is not effective enough to have the maximum queue length Qmax upper bounded by the maximum buffer capacity Cmax . This is because the first order rate ....

....Correctly 0 Send Left Rxmit Next Fig. 1. The proposed flow and error control scheme. not adjust the rate fluctuation amplitude that determines Qmax . Consequently, the first order rate control only exercises the control over bandwidth, but leaves bottleneck buffers un controlled. In [10] and [5] Qmax is analytically shown to increase with both the rate gain parameter and the connection s RTD. In [5] we developed the second order rate control, called # control, to deal with the RTD variation due to the bottleneck drift in a multicastcommunication tree. In this paper, we ....

[Article contains additional citation context not shown here]

X. Zhang, K. G. Shin, and Q. Zheng, "Integrated rate and credit feedback control for ABR services in ATM networks," in Proc. of IEEE INFOCOM, pp. 1297--1305,April 1997. Full paper version: Tech. Report, CSE-TR-35497, The University of Michigan.

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