ISSN 1068-3623, Journal of Contemporary Mathematical Analysis, 2016, Vol. 51, No. 2, pp. 103–110.


c Allerton Press, Inc., 2016.
Original Russian Text 
c B. Fathi-Vajargah, S. Ghasemalipour, 2016, published in Izvestiya NAN Armenii. Matematika, 2016, No. 2, pp. 42-53.

PROBABILITY THEORY AND MATHEMATICAL STATISTICS

Simulation of a Random Fuzzy Queuing System

with Multiple Servers
B. Fathi-Vajargah1* and S. Ghasemalipour1**
1
University of Guilan, Rasht, Iran
Received October 11, 2014

Abstract—The paper considers a queuing system that has k servers and its interarrival times and
service times are random fuzzy variables. We obtain a theorem concerning the average chance of the
event “r servers (r ≤ k) are busy at time t”, provided that all the servers work independently. We
simulate the average chance using fuzzy simulation method and obtain some results on the number
of servers that are busy. Some examples to illustrate the simulation procedure are also presented.

MSC2010 numbers : 60K05, 68U20, 65C10

DOI: 10.3103/S1068362316020060
Keywords: Multi-server queuing system; busy time; fuzzy interarrival times; average chance.

1. INTRODUCTION
Queuing systems constitute a central tool in modeling and performance of telecommunication and
computer systems. In the fuzzy case, it is assumed that the interarrival times and the service times are
random fuzzy variables. Pardoa and Fuenteb [1] proposed the analysis, development and design of a
fuzzy queuing model with a finite input source in which the arrival pattern as well as the service pattern
follow an exponential distribution with an uncertain parameter. Wanga, Liub and Watada [2] studied
a fuzzy random renewal process in which the interarrival times are assumed to be independent and
identically distributed fuzzy random variables, and two case studies of queuing systems are provided
to illustrate the application of the fuzzy random elementary renewal theorem. Wu [3] has proposed the
fuzzy arrival rate and fuzzy service rate in a queuing system. The nonhomogeneous Poisson process
with fuzzy intensity function is taken as the arrival process for this queuing system. Computational
procedures for performing simulation in the a-level sense and for obtaining the a-level closed intervals
of the system performance measure are also proposed to tackle this kind of model. Chen [4] has proposed
a procedure for constructing the membership functions of the performance measures in finite-capacity
queuing systems with the arrival rate and service rate being fuzzy numbers. Kreimer [5] studied a
real-time multi-server system with homogeneous servers (such as unmanned air vehicles or machine
controllers) and several nonidentical channels (such as surveillance regions or assembly lines), working
under maximum load regime. Zhao, Li and Huang [6] developed a queue-based interval-fuzzy electric-
power system (QIF-ESP) model through coupling fuzzy queue (FQ) theory with interval-parameter
programming (IPP). Yang and Chang [7] investigated the F -policy queue using fuzzy parameters, in
which the arrival rate, the service rate, and the start-up rate are all fuzzy numbers. The F -policy deals
with the control of arrivals in a queuing system, in which the server requires a start-up time before
allowing customers to enter.
In this paper, we simulate the average chance of the event “all the k servers are busy at time t” and
the queuing system has k servers. In other words, we estimate the average chance of the event “r servers
(r ≤ k) are busy at time t”, when all the servers work independently and the interarrival times and the
service times are random fuzzy variables. We obtain some results about the relationship between the
number of servers and busy times and idle times.
*
E-mail: [email protected]
**
E-mail: [email protected]

103
104 FATHI-VAJARGAH, GHASEMALIPOUR

The paper is structured as follows. In Section 2, we discuss the concepts and essential properties
of fuzzy set theory, fuzzy variables, random fuzzy variables and the average chance. In Section 3, we
illustrate the random fuzzy queuing system with multiple servers and estimate the average chance of the
event “r servers (r ≤ k) are busy at time t”. In Section 4, we consider the fuzzy simulation method. In
Section 5 we provide some numerical examples.

2. DEFINITIONS AND PRELIMINARIES

Credibility theory, introduced by Liu (see [8]), is a branch of mathematics for studying the behavior of
fuzzy phenomena. In this section, we introduce the basic notions of credibility theory, such us credibility
measure, credibility space, fuzzy variable, membership function, credibility distribution, expected value,
random fuzzy variable and its expected value, independence and identical distribution.
Let Θ be a nonempty set, and P be the power set of Θ, that is, the largest σ-algebra over Θ. Each
element of P is called an event. In order to give an axiomatic definition of credibility, it is necessary to
assign to each event A a number Cr{A} which indicates the credibility that A will occur.
Definition 2.1. (Liu [8]) A set function Cr defined on Θ is called a credibility measure if it satisfies the
following axioms:
Axiom 1. (Normality): Cr{Θ} = 1.
Axiom 2. (Monotonicity): Cr{A} ≤ Cr{B} for A ⊂ B.
Axiom 3. (Self-Duality): Cr{A} + Cr{Ac } = 1 for any event A.
Axiom 4. (Maximality): C{∪i Ai } = supi Cr{Ai } for any events {Ai } with supi Cr{Ai } < 0.5.
Then the triplet (Θ, P, Cr) is called a credibility space. The product credibility measure can be defined
in multiple ways. We accept the following axiom.
Axiom 5. (Product Credibility Axiom): Let Θk be nonempty sets on which Crk are credibility
measures for k = 1, 2, ..., n, respectively, and let Θ = Θ1 × Θ2 × ... × Θn . Then
Cr{(θ1 , θ2 , ..., θn )} = Cr1 {θ1 } ∧ Cr2 {θ2 } ∧ ... ∧ Crn {θn } (2.1)
for each (θ1 , θ2 , ..., θn ) ∈ Θ.
Let (θk , Pk , Crk ), k = 1, 2, ..., n, be credibility spaces, Θ = Θ1 × Θ2 × ... × Θn and Cr = Cr1 ∧
Cr2 ∧ ... ∧ Crn . Then the triplet (Θ, P, Cr) is called the product credibility space of (θk , Pk , Crk ),
k = 1, 2, ..., n.
Definition 2.2. A fuzzy variable is defined to be any real-valued measurable function defined on a
credibility space (Θ, P, Cr).
Definition 2.3. Let ξ be a fuzzy variable defined on the credibility space (Θ, P, Cr). Then the member-
ship function μ of ξ is defined by
μ(x) = (2Cr{ξ = x}) ∧ 1, x ∈ . (2.2)
Definition 2.4. Let ξ be a fuzzy variable defined on the credibility space (Θ, P (Θ), Cr), and let α ∈ (0, 1].
Then
ξα = inf{r|μξ (r) ≥ α}, ξα = sup{r|μξ (r) ≥ α} (2.3)
are called α-pessimistic value and α-optimistic value of ξ, respectively.
Definition 2.5. Let ξ be a fuzzy variable defined on the credibility space (Θ, P (Θ), Cr). Then the
expected value of ξ is defined by
 ∞  0
E[ξ] = Cr{ξ ≥ r}dr − Cr{ξ ≤ r}dr, (2.4)
0 −∞

provided that at least one of the two integrals in (2.4) is finite (see [9]).
Proposition 2.1. Let ξ be a fuzzy variable defined on the credibility space (Θ, P (Θ), Cr). Then we have

1 1 
E[ξ] = [ξ + ξα ]dα. (2.5)
2 0 α

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

 SIMULATION OF A RANDOM FUZZY QUEUING SYSTEM 105

Proof. Let ξ be normalized, that is, there exists a real number r0 such that μξ (r0 ) = 1. If r0 > 0, then in
view of (2.4), we have
 +∞  r0 
1 1 1 
E[ξ] = [r0 + Cr(ξ ≥ r)dr + r0 − Cr(ξ ≤ r)dr] = (ξα + ξα )dα,
2 r0 −∞ 2 0

implying (2.5). The case r0 ≤ 0 can be treated similarly.

Definition 2.6. The fuzzy variables ξ1 , ξ2 , ..., ξm are called independent if
i=1 {ξi ∈ Bi }} = min Cr{ξi ∈ Bi }
Cr{∩m (2.6)
1≤i≤m

for any sets B1 , B2 , ..., Bm ∈ .

Definition 2.7. A random fuzzy variable is defined to be any function from the credibility space
(Θ, P, Cr) to the set of random variables.
Definition 2.8. The expected value of a random fuzzy variable ξ is defined by
 ∞  0
E[ξ] = Cr{θ ∈ Θ|E[ξ(θ)] ≥ r}dr − Cr{θ ∈ Θ|E[ξ(θ)] ≤ r}dr. (2.7)
0 −∞

Proposition 2.2. Let ξ be a random fuzzy variable defined on the credibility space (Θ, P, Cr). Then for
any θ ∈ Θ the expected value E[ξ(θ)] is a fuzzy variable, provided that E[ξ(θ)] is finite for a fixed θ ∈ Θ.
Definition 2.9. The random fuzzy variables ξ and η are said to be identically distributed if
sup inf {P r{ξ(θ) ∈ B}} = sup inf {P r{η(θ) ∈ B}} (2.8)
Cr{A}≥α θ∈A Cr{A}≥α θ∈A

for any α ∈ (0, 1] and any Borel set B of real numbers.

Definition 2.10. Random fuzzy variables ξi , i = 1, ..., n, are said to be independent if
(1) ξi (θ), i = 1, ..., n, are independent random variables for each θ ∈ Θ.
(2) E[ξi (·)], i = 1, ..., n, are independent fuzzy variables.
Definition 2.11. Let ξ be a random fuzzy variable defined on the credibility space (Θ, P (Θ), Cr). Then
the average chance of the random fuzzy event ξ ≤ 0 is defined as
 1
Ch{ξ ≤ 0} = Cr{θ ∈ Θ|P r{ξ(θ) ≤ 0} ≥ p}dp. (2.9)
0

3. RANDOM FUZZY QUEUING SYSTEMS WITH k SERVERS

In this section we study a model of queuing system with k servers, denoted by RF/RF/k/F CF S/∞/∞,
where RF/RF means that the interarrival times and the service times are random fuzzy variables,
F CF S means that the queue discipline is “first come, first served”, and the size of source population
is infinite.
We assume that the interarrival times of customers arriving at the server are independent and
identically distributed random fuzzy variables, ξi ∼ EXP (λi ), where λi are fuzzy variables defined on
the credibility space (Θi , P (Θi ), Cri ), i = 1, 2, ..., and the service times are independent and identically
distributed random fuzzy variables, ηi ∼ EXP (μi ), where μi are fuzzy variables defined on the credibility
space (Γi , P (Γi ), Cri ), i = 1, 2, .., and ξi and ηi are independent.
For the model RF/RF/k/F CF S/∞/∞ we describe the limit (as t → ∞) of the average chance of
the event "the random fuzzy queuing system is busy at time t", when the queuing system has k servers.
The case of different number of servers is also discussed. Note that in the special case where the model
involves only one server (k = 1), this problem has been considered in [10].
Define P(t)=Pr{all of kservers are busy at time t}, and Pi (t)=Pr{the ith server is busy at t},
and observe that P(t) and Pi (t) are fuzzy variables, Pα 0 and Pα0 are the α0 -pessimistic values and the
α0 -optimistic values of P (t), respectively, and E[ μλ ] < 1.
Lemma 3.1. ([10]) Assume that in a random fuzzy queuing system RF/RF/k/F CF S/∞/∞, the
fuzzy variable λ has the same α0 -pessimistic values and the α0 -optimistic values λi , and the

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

106 FATHI-VAJARGAH, GHASEMALIPOUR

fuzzy variable μ has the same α0 -pessimistic values and the α0 -optimistic values μi , and are
continuous at the point α0 ∈ [0, 1]. Also, let the k servers work independently, then we have

 λα  λα
lim Piα (t) = and lim Piα (t) = . (3.1)
t→∞ μα t→∞ μα
Theorem 3.1. Let in a random fuzzy queuing system RF/RF/k/F CF S/∞/∞, the distributions
of ξi (θ) and ηi (γ) be non-lattice, and let the fuzzy variables λi and μi , i = 1, 2..., be continuous at
the point α ∈ (0, 1]. Also, let the k servers work independently, then we have
lim Ch{all of k servers are busy at time t} = (E[λ/μ])k . (3.2)
t→∞

Proof. From Definition 2.10 and Proposition 2.1, for ith server, i = 1, 2, ..., k, we have
 1
Ch{the ith server is busy at time t} = Cr{θ ∈ Θ|Pi (t)(θ) ≥ p}dp
0

 ∞  1
1  
= Cr{θ ∈ Θ|Pi (t)(θ) ≥ p}dpE[Pi (t)] = (Piα (t) + Piα (t))dp.
0 2 0
It follows from the definition of the limit that there exist two non-negative real numbers t1 and t2 ,
 (t) ≤ λ /μ , and for any t ≥ t 0 ≤ P  (t) ≤ λ /μ . Therefore, for
such that for any t ≥ t1 , 0 ≤ Piα α α 2 iα α α
any t ≥ max(t1 , t2 ), we have

  λα λα
0 ≤ Piα (t) + Piα (t) ≤ 2 + + .
μα μα
λα λ
Since E[ μλ ] is finite, 2 + μ + α
μα is an integrable function of α. Hence, we can apply Fatou’s lemma, to
α
conclude that
 1  1
   
lim inf (Piα (t) + Piα (t))dα ≥ lim inf (Piα (t) + Piα (t))dα,
t→∞ 0 0 t→∞

and
 1  1
   
lim sup (Piα (t) + Piα (t))dα ≤ lim sup (Piα (t) + Piα (t))dα.
t→∞ 0 0 t→∞

Next, since λα , λα , μα , μα are almost surely continuous at the point α, by Lemma 2.1 we have
 1
1  
lim Ch{the ith server is busy at time t} = lim (Piα (t) + Piα (t))dp
t→∞ 2 t→∞ 0
  1  
1 1
1 λα λ
= lim (P  (t) + 
Piα (t))dp = + α dα = E[λ/μ].
2 0 t→∞ iα 2 0 μα μα
Now, for k servers that work independently and are identically distributed, we have

k
lim Ch{all of k servers are busy at time t} = E[λ/μ] = (E[λ/μ])k ,
t→∞
i=1

and the result follows. Theorem 3.1 is proved.

Remark 3.1. If X is the number of servers that are busy at time t, then X is a random variable with
binomial distribution, that is, X ∼ B(k, Ch), where k is the number of servers and Ch is the average
chance of the servers that are busy at time t. So, the average chance of the r servers (r ≤ k) that are
busy at time t is given by
lim Ch{r servers are busy at time t} = (kr )(E[λ/μ])r (1 − E[λ/μ])k−r , (3.3)
t→∞

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

 SIMULATION OF A RANDOM FUZZY QUEUING SYSTEM 107

for r = 0, 1, 2, ..., k. Also, we have


k
lim Ch{at least r servers are busy at time t} = (ki )(E[λ/μ])i (1 − E[λ/μ])k−i , (3.4)
t→∞
i=r

Therefore, the mean of the number of servers that are busy at time t is kE[ μλ ].

Then, the average chance of all of k servers are idle at time t is given by
lim Ch{all of k servers are idle at time t} = (1 − E[λ/μ])k . (3.5)
t→∞

4. THE FUZZY SIMULATION APPROACH

Y.Liu and B.Liu [11] designed a fuzzy simulation procedure for both discrete and continuous cases.
(a) Discrete fuzzy vector: assume that f is a function, and ξ= (ξ1 , ..., ξm ) is a discrete fuzzy vector whose
joint credibility distribution function is defined by
⎧
⎪
⎪
⎪
⎪ μ1 , u = u1
⎪
⎪
⎨μ , u=u
2 2
μξ (u) = (4.1)
⎪
⎪
⎪
⎪ ...
⎪
⎪
⎩μ , u=u ,
n n

where μu = min1≤i≤m μ(i) (ui ), u = (u1 , ..., um ) ∈ m and μ(i) are the credibility distribution functions
of ξi for i = 1, 2, ..., m.
Let ai = f (ui ). Without loss of generality, we can assume that a1 ≤ a2 ≤ ... ≤ an . Then the expected
value is given by

n
E[f (ξ)] = ai pi , (4.2)
i=1

where
pi = 1/2[∨nj=i μj − ∨n+1
j=i+1 μj ] + 1/2[∨j=1 μj − ∨j=0 μj ],
i i−1
(4.3)
for i = 1, 2, ..., n and μ0 = μn+1 = 0 .
(b) Continuous fuzzy vector: assume that ξ is a continuous fuzzy vector with a credibility distribution
function μ. In this case, the expected value can be estimated by formula (4.2).

5. EXPERIENTIAL RESULTS
In this section, we present some practical applications of the model under study, to show how the
fuzzy simulation method can be used to estimate the average chance.
Example 1. Consider an investment bank with k servers. Let the interarrival times of customers be fuzzy
variables with exponential distributions with λ = (1/2/3) in minutes, and let the service times be fuzzy
variables with exponential distributions with μ = (3/4/5) in minutes for k servers. Calculate the average
chance of the event "all of k servers are busy at time t".
We use Theorem 3.1 and the simulation method, described in Section 4, to estimate the expectation
λ
E[ μ ]. The corresponding simulation results are shown in Table 1 and Figure 1, which contain the average
chance of the event “all of k servers are busy at time t” for the number of different servers. The algorithm
for simulating follows.
The Algorithm
1. Generate the random numbers e1 (i) in the interval (1,3), and the random numbers e2 (i) in the
interval (3,5), i = 1, 2, ..., n.

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

108 FATHI-VAJARGAH, GHASEMALIPOUR
e1 (i)
2. Set xi = e2 (i) .

3. Set μ(i) = min(μ1 (i), μ2 (i)).

If xi and xj have the same values, remove xj from the list of results, and set μi = max(μi , μj ).
4. Apply formulas (4.2) and (4.3).
5. Calculate E k .
Table 1 and Figure 1 show that whenever the number of servers is bigger, the average chance of the event
“all of k servers are busy at time t” is smaller, and it tends to zero as the number of servers increases.
Notice that the results are obtained without using the α-cuts and we simulate the average chance. Also,
the simulation procedure is stable in high repetitions and it is close to the true answer and it can not be
invoked in few repetitions.

k 100 500 1000 10000 20000 30000

5 0.0318 0.0364 0.0398 0.0411 0.0414 0.0414

10 0.0012 0.0014 0.0016 0.0017 0.0018 0.0018

15 3.9172e-005 4.2981e-005 6.1174e-005 6.4359e-005 6.7551e-005 6.7551e-005

20 3.9523e-007 1.8686e-006 2.3734e-006 2.9698e-006 3.0270e-006 3.0270e-006

Table 1. The results of simulation of average chance for the number of different servers for Example 1.

Fig. 1. The convergence of the fuzzy simulation for Example 1.

Example 2. Let a bank have status customer. Also, let the interarrival times of costumers be fuzzy
variables with exponential distribution with λ = (2/3/4) in minutes, and the service times are fuzzy
variables with exponential distribution with μ = (3/4/5) in minutes for any server. We want to calculate
the average chance of the event “the number of different servers that are busy at time t”. We use Theorem
3.1, Remark 3.1 and the simulation method, described in Section 4, to estimate the average chance of
the event “r servers (r ≤ k) are busy at time t”. The corresponding results of simulation are shown in
Table 2 and in Figures 2-4.

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

 SIMULATION OF A RANDOM FUZZY QUEUING SYSTEM 109

n 500 1000 10000 20000 30000 40000

r=0 Average Chance 8.1376e-004 6.5993e-004 4.9812e-004 4.7863e-004 4.3809e-004 4.2924e-004

Error 3.8453e-004 2.3069e-004 6.8879e-005 4.9386e-005 8.8500e-006 0.0000

r=1 Average Chance 0.0164 0.0109 0.0087 0.0085 0.0083 0.0080

Error 0.0064 0.0029 6.8956e-004 4.9354e-004 2.9215e-004 0.0000

r=2 Average Chance 0.0723 0.0645 0.0617 0.0603 0.0593 0.0592

Error 0.0131 0.0053 0.0025 0.0011 1.0000e-004 0.0000

r=3 Average Chance 0.2363 0.2334 0.2339 0.2211 0.2205 0.2200

Error 0.0163 0.0134 0.0039 0.0011 5.0000e-004 0.0000

r=4 Average Chance 0.3981 0.4024 0.4079 0.4081 0.4085 0.4087

Error 0.0106 0.0063 7.6731e-004 6.0000e-004 2.0000e-004 0.0000

r=5 Average Chance 0.2487 0.2684 0.2981 0.2996 0.3015 0.3036

Error 0.0549 0.0352 0.0055 0.0040 0.0021 0.0000

Table 2. The results of simulation of average chance for the number of different servers for Example 2.

Fig. 2. The convergence of the fuzzy simulation for Example 2.

In Table 2, k is the number of iterations, r is the number of servers that are busy at time t, the "Error"
rows contain the errors of the real solutions and simulated solutions, and the "Average Chance" rows
contain the results for the average chance of the event “the r servers out of 5 servers are busy at time t”,
r = 0 − −5. Then we calculate the error of simulation. Similar to Example 1, from the results of Table 2
and Figure 2, we infer that whenever the number of iterations is getting bigger the simulation becomes
closer to the real solution, and the fuzzy simulation procedure is stable in high repetitions. The Figures 3
and 4 show the error of simulation for different number of servers (out of 5 servers) that are busy at time
t for n = 10000, 20000 and n = 30000, 40000, respectively. It is clear that the error of simulation tends
to zero for n > 40000.

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016

110 FATHI-VAJARGAH, GHASEMALIPOUR

Fig. 3. The convergence of the fuzzy simulation n = 10000, 20000 for Example 2.

Fig. 4. The convergence of the fuzzy simulation n = 30000, 40000 for Example 2.

REFERENCES
1. M. J. Pardoa and D. Fuenteb, “Optimal selection of the service rate for a finite input source fuzzy queuing
system”, Fuzzy Sets and Systems, 159, 325–342, 2008.
2. S. Wanga, Y. Liub and J. Watada, “Fuzzy random renewal process with queueing applications”, Computers
& Mathematics with Applications, 57, 1232–1248, 2009.
3. H.C. Wu, “Simulation for queuing systems under fuzziness”, Intern. J. Syst. Sci., 40, 587–600, 2009.
4. S. Chen, “Parametric nonlinear programming for analyzing fuzzy queues with finite capacity”, European
Journal of Operational Research, 157, 429–438, 2004.
5. J. Kreimer, “Real-time system with homogeneous servers and nonidentical channels in steady-state”,
Computers and Operations Research, 29, 1465–1473, 2002.
6. F. Zhao, Y.P. Li and G.H. Huang, “A queue-based interval-fuzzy programming approach for electric-power
systems planning”, International Journal of Electrical Power & Energy Systems, 47, 337–350, 2013.
7. D.Y. Yang and P.K. Chang, “A parametric programming solution to the F-policy queue with fuzzy parame-
ters”, International journal of systems science, 44, 1–9, 2013.
8. B. Liu, Uncertainty Theory, 2nd ed. (Springer-Verlag, Berlin, 2007).
9. B. Liu, “A survey of credibility theory”, Fuzzy Optimization and Decision Making, 5, 387–408, 2006.
10. Y. Ning, “Busy period of random fuzzy queuing system”, Proceedings of the Sixth International Conference
on Machine Learning and Cybernetics, Hong Kong, 19-22 August, 2007.
11. Y.K. Liu and B.D. Liu, “Expected value operator of random fuzzy variable and random fuzzy expected value
models”, Intern. J. of Uncertainty, Fuzziness and Knowledge-Based Systems, 11, 195–215, 2003.

JOURNAL OF CONTEMPORARY MATHEMATICAL ANALYSIS Vol. 51 No. 2 2016