1. Introduction
Queuing models have been widely used for performance evaluation of practical systems such as communication, transportation, and service systems. For the most part, the previous studies have addressed stationary solutions of queuing models for the optimal control of long-term system performance. However, time-dependent solutions to queuing systems are more meaningful for practical systems; a good example was presented in Griffiths et al. [
1], where the authors applied the transient solution of an
queue to model and analyze a 24-hour traffic profile on the Severn Bridge in England.
The pioneering results on the transient results were presented by Luchak [
2] and Saaty [
3]. Using a modified Bessel function, they presented the transient queue-length distribution of an
queue starting with a positive number of initial customers. Later, computationally efficient solutions were presented in many other studies [
4,
5,
6,
7,
8]. For
queues, several results have been presented. Griffiths et al. [
1,
9] and Leonenko [
10] derived transient solutions for
queues starting with a positive number of initial customers, and Baek et al. [
11] extended their results to be applied to analysis in a single busy period. Furthermore, Kapodistria et al. [
12] presented the time-dependent solutions to a linear birth/immigration-death process with binomial catastrophes. They presented general results that include a background random environment.
The first study on queues with deterministic service times was conducted by Garcia et al. [
13]. Using the matrix-analytic method, they derived the transient queue-length distribution of an
queue in a computationally efficient form. Later, the queue-length formula was extended to an
queue by Franx [
14]. Recently, extensive results on the transient solution of the
queue were presented by Baek et al. [
15]. They derived not only the queue-length probability but also the waiting time distribution in closed form.
The aforementioned studies focused only on the queues in a continuous-time domain. Time-dependent solutions to queues in a discrete-time domain are even scarcer. Parthasarathy and Sudhesh [
16] were the first to derive the transient queue-length solution to a discrete-time
queue. Employing a generating function and continued fractions, they obtained a computationally efficient solution. Kim [
17] studied the same model and presented the solution in a formal form. However, we failed to find any result related to the time-dependent solution to the discrete-time queue with a constant service time. From this perspective, we present a time-dependent queue-length solution to a
queue. To the best of our knowledge, our proposed results have not been noted in existing literature.
2. Main Results
In this section, we present the main results. Consider a discrete-time
queuing system with an infinite buffer in which the timeline is divided into intervals of equal length called time slots. Customers arrive into the system according to the Bernoulli process at a rate of
p, and the service times are assumed to be a constant of
D. We assume that arrivals and service completion can occur only at the slot boundaries. More specifically, we consider the late-arrival model [
18]. Therefore, we assume that customer arrivals can occur only in
and services can be completed only in
.
Let
be the number of new initial customers at time slot 0. We define
as the number of customers at the end of the time slot
n and with the following probability:
Because plays a major role in deriving , we first derive the probability in the following theorem.
Proof. Let us define
as the number of arrivals during the time interval
and
as the
i-th arrival epoch under the condition that
. We note that the service time is a constant
D and the system starts with
j initial customers. Therefore, conditioning on
, we have
where
is an indicator function which takes 1, if A is true, or 0, and the second equality holds because it is obvious that
.
Equation (
2) means that the system can be empty at time slot
n only when the
i-th customer among
r customers arrives before
-th time slot
under the condition that
. Since the arrival process is Bernoulli process with rate
p, we have
Next, to complete Equation (
2), we need to derive
. For
, it is not difficult to have
Using Equation (
4) in Equation (
2), we obtain
Now, to obtain Equation (
1), it is sufficient to prove that
We use mathematical induction for the proof of Equation (
6). For
, we trivially have
. We now assume that Equation (
6) holds for
. Then, for
, we have
where we use
for
.
In Equation (
8), we have
where we use
for the third equality.
Using Equations (
8) and (
9) in Equation (
7), we obtain
Therefore, we prove that Equation (
6) holds for
□.
Next, we derive . We have the following theorem.
Proof. Let us define
and
as the number of arrivals during the last busy period and the time slot at which the last busy period starts, respectively. We then have
To complete Equation (
11), we first derive the first term in the second equality. We again note that the system starts with
j customers, and the service time is a constant
D. Therefore, if the last busy period starts before the
-th time slot,
customers among newly arrived customers should be served during the last busy period. Then, to become
under this situation,
customers should arrive in the system, and we have
In Equation (
13), we have
because
customers among the newly arrived customers should be served during the last busy period, if the busy period starts in the time interval
.
Furthermore, we note that
Next, we need to derive
in Equation (
11). When
, it is not difficult to have
because no service can be completed during the time interval
.
In Equation (
17), we assume that
, if
. Next, we consider the case with
. Applying a similar approach used in Equations (
12)–(
16), we have
Using Equations (
12) and (
16)–(
18) in Equation (
11), we can obtain the simplified form as
From Equation (
1), we have
Then, using the above equation in Equation (
19), we have
We then have the following simplified form:
Using algebra, we can now obtain Equation (
10) to complete the proof. □
3. Numerical Examples
In this section, we show the numerical results. We use Equations (
1) and (
10) to compute the transient probabilities
,
,
and
.
Figure 1 shows the computation results of the probabilities.
Figure 1a shows the transient probabilities under the condition that
and
(i.e.,
). Let
be the stationary probability of the
queue. If
, we can use the result in Gravey et al. [
19] to compute the stationary probability and obtain
,
,
, and
. Then, in
Figure 1a, we can confirm that each transient queue-length probability converges to the stationary value.
Figure 1b shows the transient probabilities under the condition that
and
(i.e.,
). When
, all the probabilities should be 0 as
n increases; we can confirm this in
Figure 1b.