Design Of A Fuzzy Usage Parameter Controller ... - Semantic Scholar

Design of a Fuzzy Usage Parameter Controller(FUPC) for ATM Networks Chung-Ju Chang, Chih-Feng Juan, Yung-Chih Lin, AND Ray-Guang Cheng Department of Communication Engineering National Chiao Tung University Hsinchu, TAIWAN 300 Republic of China Abstract

taking appropriate actions [l].

Thas paper presents the desagn of a fuzzy usage parameter controller (FUPC) for ATM (asynchronous transfer mode) networks The usage parameter control ( U P C ) as here based on a dual leaky bucket ( D L B ) mechanasm, wath parameters of token generatzon rate and token pool saze. I n FUPC, we further e m p l o y parameters of token-change rate and anterarnval tame of zncomzng cells as znput languastac varaables to reflect the usage satuataon an advance. Samulataon results reveal that the proposed F U P C possesses shorter response tame to detect vzolataon cells, keeps the output rate of the user close to the declared usage parameters, and thus achzeves more precase control than DLB.

A good UPC mechanism should have capabilities of detection of any non-compliant traffic situation, selectivity over the range of checked parameters, rapid response time to detect violation cells, and simplicity of implementation. UPC schemes such as jumping window, triggered jumping window, moving window, exponentially weighted moving average, and leaky bucket mechanisms were proposed [2, 31. Among these schemes, the leaky bucket mechanism was a better UPC scheme [2], and a dual leaky bucket (DLB) mechanism can control both the mean cell rate and the peak cell rate [3]. Usually, a conventional leaky bucket UPC mechanism can not detect and handle violation cells until the token pool is empty. But, based on our observation of DLB, we found that violation cells would result in a faster token-change rate

Keywords: ATM, usage parameter control, dual leaky bucket. fuzzy logic control.

than complied cells. However, it is unclear to determine how the correlation between the token-change rate and the

I. INTRODUCTION

dropping rate of violation cells would be. Fuzzy logic control ATM (asynchronous transfer mode) is one of the key technologies for handling multimedia traffic in highspeed networks. In order to guarantee the quality of service

FLC synthesizes human experience and intuition into control rules, which can carry through adaptive control actions.

(QoS) of existing calls while maximizing network utilization, two related control mechanisms have been identified: connection admission control (CAC) and usage parameter control (UPC). CAC is defined as “the set of actions taken by the network during the call set-up phase in order t o determine whether a connection request can be accepted or should be rejected, ” while UPC is defined as “the set of actions taken by the network to monitor and control traffic, in terms of traffic offered and validity of the ATM connection, at the end-system access” [l]. The main purpose of UPC is to protect network resources from malicious and unintentional misbehavior, which may affect the QoS of other existing connec-

In recent years, FLC is widely employed to deal with traffic control problems of ATM networks [4, 5, 6, 71. In [4], a fuzzy traffic controller was proposed that simultaneously incorporates CAC and congestion control. Bonde and Ghosh [5] introduced fuzzy mathematics to provide flexible and high-performance solution t o queue management in ATM networks. Ndousse [6] proposed a fuzzy logic implementation of the leaky bucket mechanism that used a channel utilization feedback via QoS parameters to improve performance. Catania, Ficili, Palazzo and Panno [7] proposed a fuzzy policer to detect violations of the parameter negotiated. The fuzzy policer is a window-based control mechanism

tions, by detecting violations of negotiated parameters and

$ 1 0.00 0 1 997 IEEE 0-7803-3925-8/97

(FLC) is a suitable mechanism to cope with this problem.

21 5

in which the maximum number of cells that can be accepted

LB at the present time. And denote Ci , Ci to be the num-

in the i-th window of length T is dynamically updated by a fuzzy logic controller. Comparative studies showed that

bers of tokens in the token pools for peak-rate-controlled LB and mean-rate-controlled LB at time tl. A C , and AC2

the proposed fuzzy approaches significantly improved system performance compared with conventional approaches.

are long-term observation of token-change rates of peak-rate-

In this paper, we propose a fuzzy usage parameter con-

controlled and mean-rate-controlled LBs; they are the summations of token-change rates a t instants when each cell of

troller (FUPC) to police violation cells using a fuzzy dual

the connection arrives. At is the interarrival time between in-

leaky bucket (FDLB) . The FDLB has two leaky buckets (LB). One is for mean-rate violation control and the other for peak-rate violation control. Both LBs have their own to-

coming cells, which can indicates user’s short-term dynamic behavior. The operation of the token-change rate estimator is listed below.

ken pools and leaky rates. For a connection, we employ the

FUNCTION token-change rate estimator()

token-change rate of the dual leaky bucket (DLB) and the interarrival time of incoming cell as input linguistic variables for FDLB. With bell-shaped membership function, we fuzzify these inputs into fuzzy parameters and generate a crisp out-

BEGIN F U N COMPUTE()

At = t - tl; t is the present time A C , = A C i + (C1 - Ci) AC2 = AC2 (C2 - Ci) ti =ti At

put value t o police violation cells. Simulation results show that the proposed FUPC can possess shorter response time,

+

keep the user’s output rate close t o declared usage parameters, and thus achieve more precise control than the conven-

+

End COMPUTE sent AC,, AC2, and At to fuzzy DLB.

tional DLB.

UPDATE( ) return A C l , AC2, and End UPDATE()

11. THEFUZZY USAGE PARAMETER CONTROLLER

ti

t o policing parameter regist

END FUN

The proposed FUPC is shown in Fig. 1 which consists of a VPI/VCI processor, a policing parameter register, a policing parameter estimator, a token-change rate estimator, and an FDLB controller. At user-network-interface (UNI) of an ATM network, each user’s message is packetized into fixed-length ATM cell. When a cell arrives at the FUPC, the VPI/VCI processor uses the VPI/VCI of the cell as index to read parameters stored in policing parameter register. The parameters are: token generation rates for peak-rate and mean-rate leaky buckets, denoted by R1 and

Rz, respectively; arrival time of the latest cell of the virtual

Afterwards, the token-change rate estimator feeds A C l , AC2, and At into the FDLB which is an FLC implemention of the DLB. Details about FLC can be found in [SI. With the input variables ( A C l , ACz, A t ) , FDLB generates a control action

2 , which

provides an effective control signal

for pass/drop processor. According to the control signal x , the pass/drop processor determines the policing function of the incoming cell of this connection.

111. FUZZY DUALLEAKYBUCKETCONTROLLER

channel connection (VCC) or virtual path connection (VPC), denoted by tl; and accumulation of token-change rates for peak-rate and mean-rate leaky bucket, denoted by AC1 and AC2, respectively. R1 and RZ are policing parameters generated by policing parameter estimator from traffic parameters of peak cell rate R p , mean cell rate R,, and mean burst

We choose A C l , AC2, and At as input linguistic variables for the FDLB controller, where ( A C l ,AC2) accounts for the long-term observation of token-change rate when a cell arrives and At presents as a short-term observation of the cell’s interarrival time. The output linguistic variable z is an indication of the degree of violation/compliance

length Lg, while tl, A C l , and AC2 are parameters coming from token-change rate estimator.

for an incoming cell.

The token-change rate estimator is used to update

For the peak-rate-controller LB, the term set is defined as T ( A C 1 ) ={High ( H i l ) , Low ( L o l ) } . For mean-

AC1, AC2, and compute interarrival time between cells A t . Denote ClrC2 to be the numbers of tokens in the token

rate-controlled LB, the term set is T(ACz)={High ( H i z ) , Low ( L o z ) } . The term set of the interarrival time At is

pools for peak-rate-controlled LB and mean-rate-controlled

21 6

T(At)=:{Long ( L g ) , Short (Sh)}. And, the term set of x is defined as T(x)={Violated ( V), Weakly Violated ( W V ) , Weakly Complied (W C ) ,Complied (C)}.

Table 1: The rule structure for the fuzzv DLB controller

In this paper, bell-shaped membership functions were employed to define the terms of the term sets used in the FDLB. A bell-shaped function f(z; m, U ) and one-side bellshaped functions g(z; m , U ) and h ( z ;m, U ) are used and expressed as

f(x; m, U ) = exp(-

(x - m)2 62

1 The membership functions associated with terms V, WV,

g(x; m’

=

{

ezp(-

WC, and C in T ( x ) , denoted by p v , p w v , p w c , and p c ,

-for x 5 m otherwise

(2)

otherwise for m s x

(3)

w) 1

respectively, are given by

and

h(x;m , U ) =

e z p ( - w )

where m is the mean and U is the variance of the bell-shaped function. In the application, the two parameters m and U of membership functions for A C l , AC,, and At are initially specified according to our basic knowledge about DLB. The set of the membership functions associated with

Sugeno [9] suggested four methods to develop the con-

terms in the term set of AC,, is denoted by M(AC1).

trol knowledge base of an FLC, including “Obtained from

M(aCl)={p~ip ~ , ~ ~where ~ } p, ~ and i p~~~ ~ are the membership functions for Hi1 and Lol, respectively, and are given

Experience and Knowledge,” “Modeling the Operator’s Control Actions,” “Modeling a process,” and “Self Organization.” From our experience and knowledge on DLB, the rules are initially set in Table 1. These rules are determined by

by

our observation t o a UPC system with DLB controller. We and

take Rule 4 as an example to describe the decision process at the first stage below.

Rule 4: IF AC, is Low, AC, is High, and At is Long, then

p~~~ Similarby, the membership functions M ( A C ~ ) = { ~, H ~ ~}

is Weakly Complied. ‘ACl is Low’ (‘ACZ is High’) means that the to-

2

are given by

ken change rate of peak-rate-controlled (mean-ratecontrolled) LB is slight (great) over during a long term observation. If the interarrival time ( A t ) is long under normal circumstance, the tokens of two leaky buckets should tend to increase. Therefore, we can conclude

and

that the traffic is not bursty right now and x could be

And the membership functions M(At)={psh ,pLg} are expressed as

’Weakly Complied’. Tsukamoto’s defuzzification method [8, p. 771 was used to combine all fuzzy rule’s output membership values into a crisp output value. The defuzzification equation is as follow:

21 7

Eq. (lo)-( 13). The rule structure of the FDLB was further optimize by a genetic algorithm (GA) [lo].

Table 2: Three types of violation I

violation case I violation oarameter Mean rate I Rm > R ~ o Peak rate R , > R,o Mean & Peak rate I R , > RmO,R , > R.0

I const. Darameter I

I I I

R, = R P o Rm=R,o

According t o the algorithm proposed in [3] and the declared usage parameters, we can calculate the leaky rate R1=83.0587, the token pool size K1=2 of peak-rate controlled LB, the leaky rate R2=30.1701, and the token pool

'

size 1(2=4316 of mean rate controlled LB. Fig. 2 showed the cell loss ratio PI,,, versus the meanwhere n denotes the number of rules with firing strength

rate violation ratio from R,/R,o=%

(wi)greater than 0 ,and xi is the amount of control action recommended by rule i.

the FDLB or the DLB. It was found that the behavior of

Iv. SIMULATION

AND

using

the FDLB was closer t o the ideal curve plotted by Eq. (15) than the DLB. Thus, the performance of the FDLB was better than the DLB. It was because the FDLB utilizes more information to control mean rate violation. The pre-

DISCUSSION

In the simulations, an ON-OFF model is used to model a traffic source. The peak rate RPo and the mean rate

cision improvement was about 8% at R,/R,o=1.1.

RmO are declared by a user and used t o characterize the ONOFF traffic source. During ON (talkspurt) state, cells are

Fig. 3

showed PI,,, versus peak-rate violation ratio from Rp/Rpo=l t o Rp/RPo=2. In this case, the FDLB outperforms the DLB; the precision improvement was about 15% at Rp/Rpo=l.l.

generated with peak rate (Rp); during OFF (silence) state, no cell is generated. The source has a transition rate A 0 in the OFF state and a transition rate A 1 in the ON state. Different kinds of violations, shown in Table 2, are considered, where Rp, R, are measured peak rate, mean rate, peak-rate duration, and RPolR,o

to R,/R,o=2

Fig. 4 showed Pl,,, versus the mean-rate and peak-rate violation ratios. From the simulation result, it was found that the FDLB possessed better control performance than DLB. Fig. 5 showed Pi,,, versus the time taken to detect violation

denote the declared traffic parame-

cells of the user whose mean rate was violated under the con-

ters.

dition that R,/R,o=1.5.

The penalty mechanism for violation is here described by the equation below [3, 71

In the simulation, it was assumed

that the user had already violated the usage parameter contract at the beginning of the simulation. It can be found that the FDLB required shorter response time to detect violation than the DLB did.

V. CONCLUSION

where PI,,, is the cell loss ratio, and a is the violation factor,

a. 5 1 denotes the case of no violaton and

a = &Aor a R ~ o Rpo thus PI,,, = 0; a

PI,,, = 1 -

>

In this paper, we propose a design of fuzzy usage parameter controller (FUPC) which is a fuzzy implementation of dual leaky bucket (FDLB). Simulation results showed that the FDLB possessed a precise control over DLB; the output rate was closer t o the declared rate and shorter re-

1 denotes the case of violation and

i.

From the knowledge of DLB, parameters of membership functions for input linguistic variables of the FDLB were selected as follows: mH;1=0.960343, m~,1=0.752461, a~i1=0.080343, and a~,1=0.102461 for p ~ i l ( A C 1 ) and p ~ ~ i ( A C 1in) Eq. (4) and Eq. (5); m~;2=0.946082, m~,2=0.750760, a~i2=0.096082, and

UL,~

=0.100760 for

p ~ i 2 ( A C 2 ) and P L ~ ~ ( A C Z in) Eq. (6) and Eq. (7); and msh=0.186632, m~,=0.216730, ash=0.036632, and ~ ~ . , = 0 . 0 0 6 7 3 0for /.ish(&) and p ~ ~ ( A int ) Eq. (8) and Eq. (9). Also, the parameters of the membership functions for output linguistic variables were given by mv =1.0, mwv=-0.6281, mwc=0.307714, and mc=0.609428 in

21 8

sponse time was obtained to detect violation cells. With the aid of fuzzy logic control and DSP technique, FUPC can be implemented by a small FUPC chip. We believe t h a t the FUPC could work perfectly for traffic policing in high speed networks.

ACKNOWLEDGEMENT This work was supported in part by the Computer and Communication Labs, ITRI, under contract number G4-8501213 and by the National Science Council, Taiwan, under contract number NSC 85-2213-E009-075.

References [l] The ATM forum technique committee: “Traffic management specification, version 4.0,” April 1996. (21 E. P. Rathgep, “Modeling and Performance Comparision of Policing Mechanism for ATM Networks,” IEEE Journal on Selected Areas in Commun., vo1.9, no.3, April 1991. lo-”

!

lo-’

1

[3] M. Butto’, E. Cavallero, and A. Tonietti, “Effectiveness of the ‘Leaky Bucket’ policing mechanism in ATM networks,” IEEE Journal Selected Areas in Commun., vo1.9, no.3, p p ~335-342, April 1991.

[4] R. G. Cheng and C. J . Chang, “Design of a fuzzy traffic controller for ATM networks,” IEEE/ACM Trans. Networking, vol. 4, no. 3, pp. 460-469, June 1996.

1.2

1.6

1.4

2

1.8

1

mean rate violation Figure 2: Cell loss ratio of FDLB and DLB when mean rate violated from Rm/Rmo=l t o Rm/R,o=2

[5] A. R. Bonde and S. Ghosh, “A comparative study of fuzzy versus ‘fixed’ thresholds for robust queue management in cell-switching networks,” IEEE/ACM Tran. Networking, vol. 2 , Aug. 1994.

[SI T. D. Ndousse, “FUZZY neural control of voice cells in ATM network,” IEEE J. Select. Areas Commun., vol. 12, pp. 1488-1494, Dec. 1994. 10-71

[7] V. Catania, G. Ficili, S. Palazzo, and D. Panno, “A coinparative analysis of fuzzy versus conventional policing mechanisms for ATM networks,” IEEE/ACM Trans. Networking, vol. 4, no. 3, pp. 449-459, June 1996.

I

J 1.2

1

1.4

1.8

2

1.8

peak rate violation Figure 3: Cell loss ratio of FDLB and DLB when peak rate violated from Rp/Rpo=l to Rp/Rpo=2

[8] H.-J. Zimmermann, “Fuzzy set theory and its applications,” 2nd revised edition, Kluwer Academic Publishers, pp. 11-17, 1991. [9] M. Sugeno, “An introduction survey of fuzzy control,” Injormation Science, 36:95-83, 1985. [lo] D. E. Goldberg, “Genetic algorithms,” Addison-Wesley Publishing, 1989. RI

c;

.-

1.2

~

14

16

2

1.8

mean rate and peak rate violation

Policmg R

Figure 4: Cell loss ratio of FDLB and DLB when mean rate and peak rate were violated

R.

A

I

x

g B

0.02

-

gOoq5-

3I

001

-

0 0 005

-

1u

0 290

Figure 1: The configuration of the FUPC

295

300

305

310

315

320

325

530

Time (unit : seconds)

Figure 5: Response time of FDLB and DLB when mean rate were violated

21 9