Applied Mathematical Modelling 54 (2018) 743–751

Contents lists available at ScienceDirect

Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm

Time-dependent analysis for a two-processor heterogeneous

system with time-varying arrival and service rates
Sherif I. Ammar a,b,∗, Yousef F. Alharbi a
a
Department of Mathematics, Faculty of Science, Taibah University, Medinah, Saudi Arabia
b
Department of Mathematics, Faculty of Science, Menofia University, Menofia, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we investigate the time-dependent behavior of a two-processor heteroge-
Received 6 September 2016 neous system where the arrival and service rates are allowed to vary with time. We derive
Revised 3 October 2017
an integral equation where the time-dependent probabilities of the two-processor hetero-
Accepted 17 October 2017
geneous system are expressed in terms of a Volterra equation of the second kind. The
Available online 6 November 2017
effectiveness of our procedure is illustrated with some numerical examples. Finally, a brief
Keywords: comparison is given to show the efficiency and accuracy of the proposed method.
Transient probabilities © 2017 Elsevier Inc. All rights reserved.
Two processor heterogeneous system
Queues with varying rates
Modified Bessel functions
Volterra equations

1. Introduction

It is useful to understand queueing systems with time-varying parameters as such systems are commonly seen in queue-
ing theory. Queueing systems with time varying parameters are used to model a wide variety of applications. Some specific
cases and applications of these kinds of queues are given in Collings and Stoneman [5], Green and Kolesar [8], Koopman
[9], Kolesar [14] and Landauer and Becker [18]. These applications include waiting time for emergency ambulance services,
security checkpoints, automatic teller machines, arrival and departure clearance for aircraft at airports, multi-car dispatch of
police and many others.
Much effort has been devoted to creating new tools for the analysis of queues with time-varying rates. This is because
standard methods for treating time-independent models, cannot be easily generalized to non-stationary cases. Therefore,
several authors have attempted to find transient solutions for non-stationary models by introducing novel methods. Some
important works on time-independent queues are given in Newell [25] and Keller [13]. They use the approximations and/or
perturbation methods to identify several different time ranges where the behavior of the queue is different. Markovian single
and multiple server queues with time-varying parameters have been analyzed by many authors (See e.g., [2,10–12,15,20–
22,30,32,33] and the references therein). Extra references and outline of numerous mathematical techniques concerning
queueing systems, where all the parameters depend on time, are accessible in Massey [24], Schwarz et al. [28] and Stolletz
[29].
The essential supposition in concentrating on multi-server queueing systems is that the structure of the servers homo-
geneous, i.e., the individual service is the same for all servers. In real life, the servers are diverse in their service rates. In

∗
Corresponding author at: Department of Mathematics, Faculty of Science, Taibah University, Medinah, Saudi Arabia.
E-mail address: sherifamar20 0 0@yahoo.com (S.I. Ammar).

https://doi.org/10.1016/j.apm.2017.10.021
0307-904X/© 2017 Elsevier Inc. All rights reserved.
744 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751

Fig. 1. Model of a two-processor heterogeneous system.

most cases, for the purpose of simplicity, specialists suppose homogeneity to get a simple analysis. Heterogeneous servers
are included in the models for some viable situations, such as, nodes in telecommunications networks with links of various
amplitudes, nodes in wireless systems serving diverse mobile users, servers shaped with various processors as a conse-
quence of system updates, communications network system supporting communication channels with various transmission
rates, multiprogramming computer systems that spools their yield for using a number of printers with various speeds, or
scheduling jobs on functionally equivalent processors of a local computer network. In a heterogeneous situation, resources
are independent, appropriated, thick, and dynamic. Subsequently, they ought to be viable and booked to make the best pos-
sible use of the resources. For more points of interest on this subject, see Chakka and Do [4], Isguder and Kocer [7], Larsen
and Agrawala [16] and Lin and Kumar [17].
Onother hand, a multiprocessor system is considered to be one of the most important systems in queueing systems,
because it has wide applications in telecommunications, flexible manufacturing systems, reliability and traffic control. Many
researchers have studied multiprocessor systems. In Mitrani [23], the steady state probabilities have been obtained for some
multiprocessor systems. For related literature, refer to Parthasarathy et al. [26] and Parthasarathy and Sudhesh [27], and the
references cited therein.
We consider a multiprocessor system consisting of two types of processors, which for convenience will be referred to as
the "main" and the "backup" processors. Each job requires exactly one processor for its execution. When both processors
are idle, the main processor is scheduled for service before the backup processor. In Trivedi [31], a similar model with two
processors main and backup was presented and discussed for the case of the steady state. In Dharmaraja [6], the transient
solution has been obtained for the same system where all its parameters are time-independent. Indeed, there are several
papers that dealt with a two-processor heterogeneous system when all parameters are time-independent. For instance, in
Ammar [3], the transient behavior of a two-processor heterogeneous system with catastrophes, server failures, and repairs
is investigated. Contrary to all the aforementioned works, in Stadje [30] the transition probabilities of M(t)/M(t)/2 with
homogeneous servers have been discussed when the arrival and service rates are allowed to vary with time.
Consequently, the motivation of the current work is to discuss the transient probabilities for a two-processor heteroge-
neous system where the arrival and service rates are allowed to vary with time. Using generating functions, the present
model can be reduced to solving Volterra integral equations. We can express Pn + 1, 1 (t), the probability that n + 1 jobs are in
the queue or are being served in terms of a Volterra equation of the second kind. Each of the other transient probabilities
is expressed in terms of integral equations in P11 (t).
This article is organized as follows: Section 2 gives a description of the suggested model and discusses the main assump-
tions considered. Section 3, presents a detailed analysis of the main results by using the generating function technique to
obtain the transient probabilities in terms of a Volterra equation of the second kind. Section 4 contains numerical results,
in addition to a comparison with the Runge–Kutta method, a simulation and the results in Dharmaraja [6] and Zhang and
Coyle [33]. Finally, the main conclusion of the paper is presented in Section 5.

2. Model descriptions

Consider a computer system consisting of two processors, a main processor, and a backup processor. A detailed descrip-
tion of the system is as follows:

(1) Jobs arrive at the system according to a Poisson process with an arrival rate λ(t). Service is exponentially distributed,
and two servers provide heterogeneous service rates μ1 (t) and μ2 (t) such that μ2 (t) ≤ μ1 (t).The queueing structure
is shown in Fig. 1. The state-transition diagram of the system is given in Fig. 2.
(2) Each job needs only one server for service and the jobs select the servers on the basis of fastest server first (FSF).
(3) Let QM (t) be the number of customers in this queue (including the one in service) and QB (t) ( = 0, 1) denote whether
there is anyone being serviced by the backup processor. Clearly, QM (t) can take the values {0, 1, 2, …}, and whenever
QM (t) > 1, QB (t) = 1. The total number of customers in the system is obtained by QM (t) + QB (t). Let
Pnm (t ) = P [ QM (t ) = n, QB (t ) = m].
From these assumptions, the forward equations can be written as follows:
 (t ) = −λ(t )P (t ) + μ (t )P (t ) + μ (t )P (t ),
P00 (2.1)
00 1 10 2 01
 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751 745

Fig. 2. State-transition diagram for a two-processor heterogeneous system.

 (t ) = −(λ(t ) + μ (t ))P (t ) + λ(t )P (t ) + μ (t )P (t ),

P10 (2.2)
1 10 00 2 11

 (t ) = −(λ(t ) + μ (t ))P (t ) + μ (t )P (t ),
P01 (2.3)
2 01 1 11

 (t ) = −(λ(t ) + μ(t ))P (t ) + λ(t ) (P (t ) + P (t )) + μ(t ) P (t )

P11 (2.4)
11 01 10 21

and

Pn 1 (t ) = −(λ(t ) + μ(t ))Pn1 (t ) + λ(t ) Pn−1,1 (t ) + μ(t ) Pn+1,1 (t ), n ≥ 2 (2.5)

where μ(t) = μ1 (t) + μ2 (t). Note that there are i customers at time s, associated with the initial condition Pn − 1, 1 (s) = δ ni for
i > 1, since δ refers to the Kronecker symbol.

3. Transient solution analysis

Define the probability generating function P(z, t) for the transient state probabilities as
∞

P (z, t ) = r0 (t ) + Pn,1 (t )zn−1 , |z| ≤ 1 (3.1)
n=0

with

r0 (t ) = P00 (t ) + P01 (t ) + P10 (t ) and R0 (t ) = r0 (t ) + P11 (t ).

It is taken into account that the generating function satisfies the following conditions

P (z, s ) = 1, i<2
and

P (z, s ) = zi−1 , i ≥ 2.
Using the system of Eqs. (2.1)–(2.5), we obtain the following differential equation for P(z, t):
∂ P (z, t )  
− λ(t )(z − 1 ) + μ(t )(z−1 − 1 ) P (z, t )
∂t  
= − λ(t )(z − 1 ) + μ(t )(z−1 − 1 ) r0 (t ) + μ(t )(1 − z−1 )P11 (t ). (3.2)
The solution of (3.2) can be obtained as follows:-,
 t    
P (z, t ) = z (η, t ) μ(η )(1 − z−1 )P11 (η ) − λ(η )(z − 1 ) + μ(η )(z−1 − 1 ) r0 (η ) dη + P (z, s )z (s, t ) (3.3)
s

where
 
t

z (s, t ) = exp λ(η )(z − 1 ) + μ(η )(z−1 − 1 ) dη .

s

Using integration by parts together with the relations

r0 (t ) = −λ(t )(P01 (t ) + P10 (t )) + μ(t )P11 (t )

and
∂ z (s, t )

= − λ(s )(z − 1 ) + μ(s )(z−1 − 1 ) .z (s, t ),

∂s
746 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751

we can write the solution of (3.2) in the form

 t  
P (z, t ) = z (η, t ) μ(η )(1 − z−1 ) P11 (η ) + −λ(η )[P01 (η ) + P10 (η )] − μ(η )P11 (η ) dη
s
 t  
+ z (η, t )r0 (η )|s + P (z, s )z (s, t ) = z (η, t ) λ(η )[P01 (η ) + P10 (η )] − μ(η )P11 (t )z−1 dη
t

s
+ r0 (t ) − r0 (s )z (s, t ) + z (s, t )P (z, s ). (3.4)
Define
⎧
⎪ an+r (s, t ) br (s, t ) −a(s,t )−b(s,t )
⎨ ∞ e , n ≥ 0 for
r=0
(n + r )! r!
I˜n (s, t ) =
⎪∞ ar (s, t ) br−n (s, t ) −a(s,t )−b(s,t )
⎩ e , n < 0 for
r=0
r! ( r − n )!
where
 t  t
a(s, t ) = λ(η )dη and b(s, t ) = μ ( η )d η .
s s

The function I˜n (s, t ) is related to the modified Bessel function In (x).
Referring to [33] Zhang and Coyle, it can be shown that
∞

z (z, t ) = zn I˜n (s, t )
n=−∞

 t ∞
  
P (z, t ) = zn I˜n (s, t ) λ(η )[P01 (η ) + P10 (η )] − μ(η )P11 (t )z−1 dη + r0 (t ) + (P (z, s ) − r0 (s ))z (z, t ). (3.5)
s n=−∞

Comparing the coefficients on zn in the case where i > 1, r0 (s) = 0, and P(z, s) = zi − 1 , we obtain
 t  
Pn+1,1 (t ) = I˜n (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜n+1 (η, t )μ(η )P11 (η ) dη + I˜n−i+1 (s, t ) (3.6)
s

for n > 1 and

 t  
R0 (t ) = I˜0 (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜1 (η, t )μ(η )P11 (η ) dη + r0 (t ) + I˜−i+1 (s, t ) (3.7)
s

for n = 0. Since R0 (t) − r0 (t) = P11 (t), we can write this as

 t  
P11 (t ) = I˜0 (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜1 (η, t )μ(η )P11 (η ) dη + I˜−i+1 (s, t ). (3.8)
s

If R0 (t) = 1, then we have

 t  
Pn+1,1 (t ) = I˜n (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜n+1 (η, t )μ(η )P11 (η ) dη + P11 (s )I˜n (s, t )
s

for n > 0, and

 t  
R0 (t ) = I˜0 (η, t )λ(η )(P01 (η ) + P10 (η )) − I˜1 (η, t )μ(η )P11 (η ) dη + r0 (t ) + P11 (s )I˜0 (s, t )
s

for n = 0. We can rewrite (3.8) as

 t  
P11 (t ) = I˜0 (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜1 (η, t )μ(η )P11 (η ) dη + P11 (s )I˜0 (s, t ).
s

Combining these cases, we have

 t  
Pn+1,1 (t ) = I˜n (η, t )λ(η )[P01 (η ) + P10 (η )] − I˜n+1 (η, t )μ(η )P11 (η ) dη + P11 (s )I˜n (s, t )
s

+ (1 − R0 (s ))I˜n−i+1 (s, t ) (3.9)

for n > 0, and
 t  
P11 (t ) = I˜n (η, t )λ(η )[P01 (η ) + P10 (η )]−I˜n+1 (η, t )μ(η )P11 (η ) dη+P11 (s )I˜n (s, t )+ (1 − R0 (s ))I˜n−i+1 (s, t ) (3.10)
s

for n = 0.
 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751 747

The remaining transient probabilities P00 (t), P01 (t) and P10 (t) can be obtained by solving the system of Eqs. (2.1)–(2.3). To
do that, we rewrite the system of Eqs. (2.1)–(2.3) in the following matrix form:
dP(t )
= P(t )A(t ) + μ2 (t )P11 (ν )e1 + μ1 (t )P11 (ν )e2 (3.11)
dt
where the matrix A(t) is given as:
 
−λ(t ) λ(t ) 0
A(t ) = μ1 (t ) −λ(t ) − μ1 (t ) 0
μ2 (t ) 0 −λ(t ) − μ2 (t )

P(t ) = (P00 (t ), P10 (t ), P01 (t )), e1 = (0, 1, 0 ) and e2 = (0, 0, 1 ).

Let U(s, t) be an evolution operator such that U(t, t) = I (the identity operator) and ∂∂ t U(s, t ) = U(s, t )A(t ), then
 t
P(t ) = [μ2 (ν )P11 (ν )e1 U(ν, t ) + μ1 (ν )P11 (ν )e2 U(ν, t )] dν + P(s )U(s, t ). (3.12)
s

We can obtain a solution of U(s, t) numerically using one of the methods for solving a linear system of ordinary differ-
ential equations.
Assume that
U(s, t ) = (Ui j (s, t ))3×3 , i, j = 0, 1, 2.
Then, after considerable mathematical manipulations, Eq. (3.12) reduces to
 t
P00 (t ) = P11 (ν )[μ1 (ν )U20 (ν, t ) + μ2 (ν )U10 (ν, t )]dν + P(s )U.,0 (s, t ), (3.13)
s
 t
P10 (t ) = P11 (ν )[μ1 (ν )U21 (ν, t ) + μ2 (ν )U11 (ν, t )]dν + P(s )U.,1 (s, t ), (3.14)
s
 t
P01 (t ) = P11 (ν )[μ1 (ν )U22 (ν, t ) + μ2 (ν )U12 (ν, t )]dν + P(s )U.,2 (s, t ), (3.15)
s

where U.,k (s, t), k = 0, 1, 2 is the k column of the matrix U(s, t).
Therefore, P11 (t) can be written in terms of the Volterra equation of the second kind,
 t
 t
P11 (t ) = P11 (η ) [λ(ν )(μ1 (ν )U21 (η, ν ) + μ2 (ν )U11 (η, ν ))]dν + P(s )U.,1 (s, η )λ(η )I˜0 (η, t )
s η
 t

+ [λ(ν )(μ1 (ν )U22 (η, ν ) + μ2 (ν )U12 (η, ν ))]dν + P(s )U.,2 (s, η )I˜0 (η, t ) − μ(η )I˜1 (η, t ) dη
η

+ I˜−i+1 (η, t )(1 − R0 (t )) − R0 (s )I˜0 (s, t ). (3.16)

Thus, Eqs. (3.9), (3.13), (3.14) and (3.16) represent the complete transient solution for the two-processor heterogeneous
system.

4. Numerical illustration

Transient system size probabilities are usually difficult to obtain analytically, even in the simple case of a two-processor
heterogeneous system with Poisson arrivals, and independent exponential service times when the infinite series of Bessel
functions or their integrals are involved, see Ammar [3]. Another major difficulty is that many properties that could be
suitable for stationary systems do not suit the non-stationary systems. Here, we have considered a complex queue in which
the arrival and departure rates are allowed to vary with time. Hence, it is important to develop numerical techniques for
solving the resulting equations in order to gain insight into the behavior of various system characteristics.
An integral equation for the transient probabilities is obtained analytically. However, it is important to visualize the
solutions in practical situations. Therefore, the transient system probabilities are presented numerically in the following
subsections.

4.1. Approximation procedure of the proposed method

In this section, we describe a procedure of approximation that we use to show the performance of the proposed method.
To approximate the transient probabilities numerically, we discretize a function of the form F(s, t) into a matrix of values
whose i, jth entry is F(ti ,tj )twi,j , where t0 < t1 < t2 <  < tn = t is a partition of the interval (t0 ,t], ti + 1 − ti = t is
constant, and wi,j is a numerical integration weight. Any of the Gregory integration rules can be used here. The importance
748 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751

Fig. 3. Time dependent system size probabilities. (a) Probabilities of P00 (t), P10 (t) and P01 (t). (b) Probabilities of Pn1 (t) for n = 1,…,7.

of using a Gregory method is that the numerical integration weights are all 1 except for the first and last k weights for a
k + 1 firstorder Gregory method.
We use Simpson’s rule, which is the second-order Gregory method. For the discussion of Gregory integration methods
in the context of solving Volterra equations, see Linz [19]. For Simpson’s rule wi,j = 0.5 = w0,i , w0,0 = 0 and all other weights
are 1. The numerical integration of the product H(s, t) = ∫H(s, u)G(u, t)du is just the product of two matrices: a discretized
unweighted F(., .) and a discretized weighted G(., .). The results are represented by the unweighted discretized matrix version
of H(., .). To approximate the solution to the Volterra equation

F (s, t ) = E (s, t ) + G(s, u )F (u, t )du,

we need to compute E(I − G) − 1 , where E is the discretized matrix approximation of E(., .), I is the appropriately dimensioned
identity matrix, and G is the discretized weighted version of G(., .).
When the functions are diagonally periodic such that F(s, t) = F(s + rp, t + rp), where p is the period and r is any integer,
then it is only necessary to store a portion of these matrices.

4.2. Empirical example

In this section, we use an empirical example to explore the behavior of the proposed system. Using the obtained results,
we carried out an extensive numerical work on an IBM computer system for a variety of arrival and departure rates to
compute the transient probabilities for the system under study. Furthermore, no difficulty was encountered for any value of
the model parameters.
For a better understanding of the system behavior, we will analyze the model with time-varying rates for the two-
processor heterogeneous system. In our analysis, we will use the following arrival and service rates:-

λ(t ) = 8[1 + sin(2t )],

 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751 749

Fig. 4. Proposed approach and Runge–Kutta approach results with parameters λ(t) = 1 + sin 2t and μ1 (t) = μ2 (t) = 2.

Fig. 5. Proposed approach and Zhang and Coyle approach results with parameters λ(t) = 3[1 + sin 2t] and μ1 (t) = μ2 (t) = 2.

μ1 (t ) = 6[1 + cos(2t )],

μ2 (t ) = 5[1 + cos(2t )]. (4.1)
In Fig. 3, the transient system size probabilities are drawn for the parameters given in (4.1) with the assumption that
initially there are no customers in the system.

4.3 Comparison. results

To confirm the efficiency and accuracy of the presented method, three comparisons are conducted between our results
and the Runge–Kutta fourth-order method, a simulation and the methods described in Dharmaraja [6] and Zhang and Coyle
[33], where the solution in Dharmaraja [6] and Zhang and Coyle [33] are obtained analytically. This means that Dharmaraja
[6] and Zhang and Coyle [33] provide an exact solution in theory.
Therefore, the graph displayed in Fig. 4 presents a comparison between the obtained results and fourth-order Runge–
Kutta method. In such a comparison, we have considered a two-processor heterogeneous system with the following arrival
and service rates: λ(t) = 1 + sin 2t and μ1 (t) = μ2 (t) = 2. From Fig. 4, it is obvious that our results and the results of the
fourth-order Runge–Kutta method coincide with each other.
In addition, we compare our results with the previous results in Zhang and Coyle [33]. For this, we considered a two-
processor heterogeneous system with the following arrival and service rates: λ(t) = 3[1 + sin 2t] and μ1 (t) = μ2 (t) = 2. It can
be seen that in Fig. 5, our results agree with the results of the approach in Zhang and Coyle [33].
750 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751

Fig. 6. Comparison between the given two methods and the simulation withthe parameters λ(t) = 9[1 + sin 2t] and μ1 (t) = μ2 (t) = 6.

Table 1
Comparison betweenthe Dharmaraja approach and the proposed approach.

n 3 5 9

t Dharmaraja approach The proposed approach Dharmaraja approach The proposed approach Dharmaraja approach The proposed approach

0.5 0.04104756 0.04104312 0.01820237 0.01820453 0.0 0 0587018 0.0 0 0581582

1 0.01929606 0.01929251 0.01582140 0.01582928 0.003096872 0.003092759
1.5 0.01124099 0.01124953 0.01120713 0.01120147 0.004250901 0.004259113
2 0.00748381 0.00748274 0.00826398 0.00826481 0.004334151 0.004331538
2.5 0.00539282 0.00539039 0.00630627 0.00630705 0.004027322 0.004022814
3 0.00409783 0.00409159 0.00497633 0.00497983 0.003621143 0.003621457
3.5 0.00323340 0.00323568 0.00403299 0.00403195 0.0 03220 097 0.003220216
4 0.00262406 0.00262401 0.00333886 0.00333880 0.002857210 0.002857210
4.5 0.00217643 0.00217641 0.00281233 0.00281232 0.002539475 0.002539475
5 0.00183673 0.00183673 0.00240273 0.00240273 0.002264660 0.002264660

Table 2
Comparison between the two givenmethods in CPU time with parameters
λ(t) = 3[1 + sin 2t] and μ1 (t) = μ2 (t) = 2[1 + cos (2t)].
CPU time in (s)
Time P00 (t) The proposed approach Zhang and Coyle approach

1.5 0.53271 146.275 149.532

4 0.77384 146.411 149.363
6.5 0.53529 146.814 149.531
9 0.77491 146.925 149.854

To demonstrate the accuracy of the obtained results, we compared them with the results of simulation and the approach
in Zhang and Coyle [33], as shown in Fig. 6.
As seen, the proposed method’s curve follows exactly the simulation outputs, unlike the method in Zhang and Coyle [33],
which indicates that the proposed technique is an effective and valuable tool. Additionally, the proposed method is more
accurate for measuring parameters out performing the approach in Zhang and Coyle [33] especially in the context of the
Markovian queueing models in the transient analysis where the parameters are varying with time.
In Table 1, the transient probabilities of the system size Pn1 (t), n = 3, 5, 9 are shown. In our comparison with Dharmaraja
[6], we have considered a two-processor heterogeneous system with constant arrival and service rates: λ = 7, μ1 = 5 and
μ2 = 3. It is to be noted that the numerical results in Dharmaraja [6] and Zhang and Coyle [33] show a good agreement
with our results. Moreover, this reveals that the obtained results presented here, are consistent with those in Dharmaraja
[6] corresponding to the case of constant parameters.
We compared the CPU time (in seconds) for obtaining the transient solution using those methods. In Table 2, some
results have been presented for P00 (t). It can be seen that CPU time (in seconds) for our approach is less than that taken by
the approach in Zhang and Coyle [33].
We made use of the accuracy measure in Alnowibet and Perros [1] to judge the accuracy of the proposed method.
 S.I. Ammar, Y.F. Alharbi / Applied Mathematical Modelling 54 (2018) 743–751 751

Table 3
Absolute and relative errors of the Dharmaraja, Zhang and
Coyle approaches.

t Dharmaraja approach Zhang and Coyle approach

Error Rel. Error Error Rel. Error

2 0.0 0 025 0.05034 0.00643 0.60978

5 0.0 0 011 0.06812 0.00358 0.58926
8 0.0 0 098 0.01085 0.00411 0.77752
11 0.0 0 015 0.03164 0.00573 0.98534
15 0.0 0 074 0.04527 0.00347 0.87153

Table 3 summarizes the absolute error and the relative absolute error with respect to the approaches in Dharmaraja
[6] and Zhang and Coyle [33]. Table 3 shows that the proposed method provides a high level of accuracy.

5. Conclusion and future work

We presented an intensive study of a two-processor heterogeneous system with time-varying arrival and departure rates.
Applications can be found in a wide variety of real systems, including servers with different types of processors as a conse-
quence of system updates, nodes in a telecommunications network with links of different capacities, and nodes in wireless
systems serving different mobile users. Based on the analytic and numerical results, several conclusions can be drawn:
1 The transient probabilities can be obtained in terms of a Volterra equation of the second kind using the generating
function technique.
2 The proposed method gives accurate results compared with the fourth order Runge–Kutta method and matched the
simulation results. This means that our method is an effective and a valuable one.
3 The proposed method takes less time than other methods in the calculations of the transient probabilities.
4 The proposed method can be applied to various systems, whether they are time dependent or not.
5 In the future, this work will be extended to two-processor heterogeneous system and multi-processor systems in the
context of synchronous or asynchronous vacation and will be extended to multi-processor system in which some of
the servers take vacation simultaneously.

References

[1] K.A. Alnowibet, H. Perros, Nonstationary analysis of the loss queue and of queueing networks of loss queues, Eur. J. Oper. Res. 196 (2009) 1015–1030.
[2] R.O. Al Seedy, A.A. El Sherbiny, S.A. El Shehawy, S.I. Ammar, The transient solution to a time dependent single server queue with balking, Math. Sci.
34 (2009) 1–6.
[3] S.I. Ammar, Transient behavior of a two-processor heterogeneous system with catastrophes, server failures and repairs, Appl. Math. Model. 38 (2014)
2224–2234. 
[4] R. Chakka, T.V. Do, The MM Kk=1 CP Pk /GE/c/LG− queue with heterogeneous servers: steady state solution and an application to performance evalua-
tion, Perform. Eval. 64 (2007) 191–209.
[5] T. Collings, C. Stoneman, The M/M/∞ queue with varying arrival and departure rates, Oper. Res. 24 (1976) 760–773.
[6] S. Dharmaraja, Transient solution of a two-processor heterogeneous system, Math. Comput. Model. 32 (20 0 0) 1117–1123.
[7] H.O. Isguder, U.U. Kocer, Analysis of GI/M/n/n queueing system with ordered entry and no waiting line, Appl. Math. Model. 38 (2014) 1024–1032.
[8] L. Green, P. Kolesar, Testing the validity of a queueing model of police patrol, Manage. Sci. 35 (1989) 127–148.
[9] B.O. Koopman, Air-terminal queues under time-dependent conditions, Oper. Res. 20 (1972) 1089–1114.
[10] C. Knessl, Exact and asymptotic solutions to a PDE that arises in time dependent queues, Adv. Appl. Prob. 32 (20 0 0) 256–283.
[11] C. Knessl, Y. Yang, Analysis of a brownian particle moving in a time dependent drift field, Asymptot. Anal. 27 (2001) 281–319.
[12] C. Knessl, Y.P. Yang, An Exact solution for an M(t)/M(t)/1 queue with time dependent arrival and service, Queueing Syst. 40 (2002) 233–245.
[13] J.B. Keller, Time-dependent queues, SIAM Rev. 24 (1982) 410–412.
[14] P. Kolesar, Stalking the endangered CAT: a queueing analysis of congestion at automatic teller machines, Interfaces 14 (1984) 16–26.
[15] E.L. Leese, D.W. Boyd, Numerical methods of determining the transient behavior of queues with variable rates, J. Can. Oper. Res. Soc. 3 (1966) 1–13.
[16] R.L. Larsen, A.K. Agrawala, Control of a heterogeneous two-server exponential queueing system, IEEE Trans. Softw. Eng. 9 (1983) 522–526.
[17] W. Lin, P.R. Kumar, Optimal control of a queueing system with two heterogeneous servers, IEEE Trans. Automat. Contr. 29 (1984) 696–703.
[18] E.G. Landauer, L.C. Becker, Reducing waiting time at security checkpoints, Interfaces 19 (1989) 57–65.
[19] P. Linz, Analytical and Numerical Methods for Volterra Equations, SIAM, Philadelphia, PA, 1985.
[20] B.H. Margolius, A sample path analysis of the Mt /Mt /c queue, Queueing Syst. 31 (1999) 59–93.
[21] B.H. Margolius, Transient solution to time-dependent multi-server poisson queue, J. Appl. Prob. 42 (2005) 766–777.
[22] B.H. Margolius, Periodic solution to the time inhomogeneous multi server Poisson queue, Oper. Res. Lett. 35 (2007) 125–138.
[23] I. Mitrani, multiprocessor systems with reserves and preferences, Queueing Syst 2 (1987) 245–259.
[24] W.A. Massey, The analysis of queues with time varying rates for telecommunication models, Telecommun. Syst. 21 (2002) 173–204.
[25] G.F. Newell, Queues with time-dependent arrival rates (parts I–III), J. Appl. Probab. 5 (1968) 436–451 (I), 579–590 (II), 591–606 (III).
[26] P.R. Parthasarathy, S. Dharmaraja, G. Manimaran, Transient solution of multiprocessor system with state-dependent arrivals, Int. J. Comput. Math. 73
(20 0 0) 313–320.
[27] P.R. Parthasarathy, R. Sudhesh, Transient analysis of multiprocessor systems, in: Proceedings of the IEEE INDICON, India, 2005, pp. 276–279.
[28] J.A. Schwarz, G. Selinka, R. Stolletz, Performance analysis of time-dependent queueing systems: survey and classification, Omega 63 (2016) 170–189.
[29] R. Stolletz, Approximation of the non-stationary M(t)/M(t)/c(t) -queue using stationary queueing models: the stationary backlog-carryover approach,
Eur. J. Oper. Res. 190 (2008) 478–493.
[30] W. Stadje, A note on the simple queue with variable intensities and two servers, Oper. Res. Lett. 9 (1990) 45–49.
[31] K.S. Trivedi, Probability and Statistics with Reliability, Queuing and Computer Science Applications, Prentice Hall, Englewood Cliffs, NJ, USA, 1982.
[32] W. Whitt, J. Zhao, Many-server loss models with non-poisson time-varying arrivals, Nav. Res. Logist. 64 (2017) 177–202.
[33] J. Zhang, E.J. Coyle, The transient solution of time-dependent M/M/1 queues, IEEE Trans. Inf. Theory 37 (1991) 1690–1696.