![]() |
Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications
by Lakhmi C. Jain; N.M. Martin CRC Press, CRC Press LLC ISBN: 0849398045 Pub Date: 11/01/98 |
Previous | Table of Contents | Next |
Taking into consideration factors like the source traffic descriptor, the amount of current network congestion along the path of the incoming call, and Quality of Service requirements of the new and the pre-existing calls is a daunting task for any mathematical model. A number of publications have recently demonstrated the merit of fuzzy logic in dealing with such a complex setting. Four distinct areas of applications, namely Fuzzy Admission Control (e.g., [24]), Fuzzy Policing (e.g., [17]), Fuzzy Rate Control (e.g., [22]), and Fuzzy Buffer Management (e.g., [27]) have been investigated. Figure 1 shows the positions of a Fuzzy Logic Controller (FLC) that would perform any one of these controls. In this chapter we will present the use of fuzzy expert systems in detail in rate regulation and in policing.
Figure 1 Fuzzy Expert System ATM Controllers
Our objective in this part of the chapter is to propose a novel controller scheme that regulates the peak rate as well as reduces the Cell Loss Rate (CLR). Our proposed scheme works in connection with the Leaky Bucket (LB). The Leaky Bucket operates like a virtual queueing system, where each cell arrival increases the bucket size by one until a maximum value of Sth, while, at constant and regular time intervals, 1/Dr, the bucket size is decremented (Figure 2). Cells arriving to find the bucket size equal to Sth are discarded or tagged. If only the LB is used, the CLR is too high if the threshold Sth is set too low. Meanwhile, the peak rate cannot be controlled if the threshold is set to a high value. As the traffic gets bursty, it is difficult to set the best choice to meet these two conditions. The proposed system considers the propagation delay time Δb to predict the possible cell loss in the near future. If cell discarding is imminent, the source transmission rate is reduced to a level that depends on the strength of the feedback signal. A correct prediction for the feedback signal is very important. If the backpressure is overdone, the additional delay time incurred on the cells whose transmission has been postponed will be intolerable, although the cell loss ratio is very low. On the other hand, if it is underestimated, the cell loss ratio may still be excessive. In this work, we propose a hybrid mechanism that uses the Leaky Bucket as the cell loss controller and generates a backpressure signal that is sent back to the transmitting source and is the outcome of fuzzy processing of the status of the LB pseudo-queue, of an indicator of the changing rate of the whole system, and of a variable that monitors the error of previous decisions. These three parameters are fuzzified to take linguistic values like HIGH, SMALL, MEDIUM, LARGE, POSITIVE, and NEGATIVE.
Figure 2 A Leaky Bucket Controller
Figure 3 The Architecture of Feedback Rate Regulator
The architecture of our proposed control model is shown in Figure 3. Assume that a traffic source declares its peak rate as Bp and mean rate as Bm. At time Ti the source generates data with a rate Ri. This rate is regulated to ℜi by a Pre-Shaping Unit (PSU) which reduces the data rate depending on the output OTi of the fuzzy controller.
A Leaky Bucket is used as the policing function to shape the regulated traffic ℜi. We define the depletion rate of LB as Dr, Dr = 1/T. The time interval to sample the status of the LB is equal to ΔT = Ti-Ti-1, for all i, j > 0. Let DTi be the number of cells discarded between time Ti and Ti + ΔT if cells arrive at a time when Si = Sth, where Si is the current value of the LB counter.
To monitor the expected number of cells that arrive if Ri is not regulated in transmitting during ΔT, we define δTi = (Si.-.Si-1) + DTi + CTi, where CTi is the estimated number of cells stopped by the PSU from time Ti to Ti + ΔT.
In our model, we use ℜi = (1-OTi) Ri to regulate the source rate. So we define CTi = (OTi × Ri × ΔT) /424. A possible scenario works as follows: at time Ti, the source begins to transmit data at rate ℜi. This status will be maintained for ΔT seconds. In the meantime, the fuzzy logic system receives the status of LB for the time interval [Ti-1, Ti]. The fuzzy logic system is able to predict possible cell discarding in time interval [Ti+1, Ti+2]. If ΔT is not less than Δb plus the processing time of the fuzzy logic system, then we can regulate the source rate in ΔT time intervals.
We assume that ρmax is the number of cells generated by the source at its peak rate Bp during ΔT. From the system description, we get δTi <[(1- Dr/Bp) × ρmax] = δmax.
δmax is an important parameter in our prediction, because it represents the maximum possible increment in the pseudo-queue of the Leaky Bucket during the period of ΔT, if the real peak rate doesnt violate the declared peak rate. We use it as a key parameter to decide the membership functions of the fuzzy controller.
Previous | Table of Contents | Next |