THE BIRTH-AND-DEATH PROCESS

BE (CS) Lecture 28
Spring 2022 Semester ▪ Most elementary queuing models assume that the inputs
(arriving jobs) and outputs (leaving jobs) of the queuing
system occur according to the birth-and-death process.
CS-438: Computer Systems Modeling
▪ The state of the system at time t (t ≥ 0), denoted by n(t),
o number of jobs in the queuing system at time t.

Lectures 28-30 ▪ The birth-and-death (B&D) process describes

probabilistically how n(t) changes as t increases.
Birth-And-Death Processes
▪ Broadly speaking, it says that individual births and deaths
M/M/1 System occur randomly, where their mean occurrence rates
M/G/1 System depend only upon the current state of the system.
Dr Syed Zaffar Qasim
Assistant Professor (CIS)
1 2

THE BIRTH-AND-DEATH PROCESS THE BIRTH-AND-DEATH PROCESS

▪ More precisely, the assumptions of the birth-and-death
process are the following:-
o Assumption 1: Given N(t) = n, the current probability
distribution of the remaining time until the next birth
(arrival) is exponential with parameter n (n = 0, 1, 2,
…).
o Assumption 2: Given N(t) = n, the current probability
distribution of the remaining time until the next death
(service completion) is exponential with parameter µn
(n = 1, 2, …).
o Assumption 3: The random variable of assumption 1
(the remaining time until the next birth) and the
random variable of assumption 2 (the remaining time
until the next death) are mutually independent.
3 4
▪ The next transition in the state of the process is either
o n → n + 1 (a single birth)
or
o n → n - 1 (a single death),
o depending on whether the former or latter random
variable is smaller.
▪ Because of these assumptions, the B&D process is a
special type of continuous time Markov chain.
▪ Queuing models that can be represented by a
continuous time Markov chain are far more tractable
analytically than any other.

CSM (CS-438) 1
 THE BIRTH-AND-DEATH PROCESS
▪ We can summarize these assumptions by the rate
diagram shown in Fig. 1.
Fig 1: Rate diagram for the birth-and-death process
▪ The arrows in this diagram show the only possible
transitions in the state of the system (as specified by
assumption 3).
▪ The entry for each arrow gives the mean rate for that
transition (as specified by assumptions 1 and 2) when
the system is in the state at the base of the arrow.
5 6
Generalized Poisson Queuing Model
▪ Consider any particular state of the system n (n = 0, 1, 2, …).
▪ Starting at time 0, suppose that a count is made of the
number of times that the process enters this state and the
number of times it leaves this state, as denoted below:-
o En(t) = number of times that process enters state n by time t.
o Ln(t) = number of times that process leaves state n by time t.
▪ Because the two types of events (entering and leaving)
must alternate,
7 8
CSM (CS-438)
Generalized Poisson Queuing Model
▪ Except for a few special cases, analysis of the birth-and-
death processes is very difficult when the system is in a
transient condition.
▪ Some results about the probability distribution of N(t)
have been obtained, but they are too complicated to be of
much practical use.
▪ On the other hand, it is relatively straightforward to
derive this distribution from the rate diagram after the
system has reached a steady-state condition.
▪ Here we develop a general queuing model
o that combines both arrivals and departures based on
the Poisson assumptions –
o i.e., the inter-arrival and service times follow the
exponential distribution.
Generalized Poisson Queuing Model
▪ Dividing through both sides by t and then letting t → 
gives
▪ Dividing En(t) and Ln(t) by t gives the actual rate
(number of events per unit time) at which these two
kinds of events have occurred, and
▪ letting t →  then gives the mean rate (expected number
of events per unit time):-
2
 Generalized Poisson Queuing Model Generalized Poisson Queuing Model
▪ To illustrate a balance equation, consider state 0.
▪ These results yield the following key principle:-
Rate In = Rate Out Principle ▪ The process enters this state only from state 1.

Fig 1: Rate diagram for the birth-and-death process

▪ For any state of the system n (n = 0, 1, 2, …),
o mean entering rate = mean leaving rate.
▪ Thus, the steady-state probability of being in state 1
▪ The equation expressing this principle is called the (P1) represents the proportion of time that it would
balance equation for state n.
be possible for the process to enter state 0.
▪ After constructing the balance equations for all the
states in terms of the unknown Pn probabilities, ▪ Given that the process is in state 1, the mean rate of
o we can solve this system of equations (plus an entering state 0 is µ1.
equation that sum of probabilities is 1) to find Pn.
9 10

Generalized Poisson Queuing Model Generalized Poisson Queuing Model

▪ For every other state there are two possible transitions

both into and out of the state.
▪ Therefore, each side of the balance equations for these
states represents the sum of the mean rates for the two
transitions involved.
▪ (In other words, for each cumulative unit of time that the
process spends in state 1, the expected number of times ▪ Otherwise, the reasoning is just the same as for state 0.
that it would leave state 1 to enter state 0 is µ1.) ▪ These balance equations are summarized in Table 1.
▪ From any other state, this mean rate is 0.
▪ Therefore, the overall mean entering rate to state 0 is µ1P1.
▪ By the same reasoning, the mean leaving rate to state 0 is
0P0.
▪ Hence the balance equation for state 0 is
µ1P1 = 0P0 11 12

CSM (CS-438) 3
 Generalized Poisson Queuing Model Generalized Poisson Queuing Model
▪ Applying this procedure yields the following results:-
▪ Notice that
o the first balance equation contains two variables (P0 and P1),
o the first two equations contain 3 variables, and so on,
o so that there always is one “extra” variable.
▪ Therefore, the procedure is to solve these equations in terms of
one of the variables, the most convenient one being P0.
▪ At the end, the requirement that the sum of all the probabilities
equal 1 can be used to evaluate P0.
Table 1: Balance equations for the birth-and-death process

13 14

Generalized Poisson Queuing Model Generalized Poisson Queuing Model

▪ To simplify notation, let ▪ When a queuing model is based on the birth-and-death
process,
o so the state of the system n represents the number of
▪ and then define Cn = 1 for n = 0. Thus, the steady-state jobs in the queuing system,
probabilities are o the key measures of performance for the queuing
Pn = CnP0, for n = 0, 1, 2, … system (L, Lq, W, and Wq) can be obtained immediately
▪ The requirement that after calculating the Pn.
▪ The definitions of L and Lq specify that
▪ implies that

▪ so that ▪ Furthermore, the relationships yield

15 16

CSM (CS-438) 4
 Generalized Poisson Queuing Model
▪ Where  is the average arrival rate over the long run.
▪ Several of the expressions just given involve
summations with an infinite number of terms.
▪ Fortunately, these summations have analytic solutions
for a number of interesting special cases, as seen later.
▪ Otherwise, they can be approximated by summing a
finite number of terms on a computer.
▪ These steady-state results have been derived under the
assumption that n and µn have values such that the
process actually can reach a steady-state condition.
▪ It also always holds when  and µ are defined and  =
/(sµ) < 1.
Kendall notation
Lecture 29
▪ A notation for the parameters of a queuing
system due for the most part to Kendall (1953)
illustrated below.
CSM (CS-438)
Generalized Poisson Queuing Model
PROBLEM
▪ A medium-sized university library provides a search
facility for its students and faculty through a single-server
database system. The read-only transactions arrive in the
system according to a Poisson distribution at the mean
rate of 15 per minute whereas the completion time is
exponential with mean value 0.05 minute. The maximum
length of queue is 3 beyond which no more transactions
are accepted.
1. Determine the steady-state probability pn of n
transactions in the system.
2. Also evaluate the key measures of performance for
the queuing system.
18
17
Kendall notation
▪ The service discipline (Z) for ordering jobs in the queue
o FIFO, LIFO, priority ordered, random ordered, and
others.
▪ The notation M/D/2///FCFS indicates a queuing
process
o with exponential interarrival times, deterministic
service times,
o two parallel servers, no restriction on the maximum
jobs allowed in the system and FCFS queue
discipline.
▪ In many situation, only the first three symbols are
used.
19 20
5
 Kendall notation The M/M/s Model
▪ Thus, M/D/2 would be a queuing system with ▪ Assumes that
o exponential input, o all inter-arrival times are independently and identically
o deterministic service, distributed according to an exponential distribution
(i.e., the input process is Poisson),
o two servers,
o that all service times are independent and identically
o no limit on system capacity distributed according to another exponential
o no limit on calling population and distribution, and
o that the number of servers is s (any positive integer).
o FIFO discipline.
▪ Consequently, this model is just the special case of the
Problem
birth-and-death process where
In an M/M/5 system, jobs arrive at the rate of 24 o the queuing system’s mean arrival rate and mean service
jobs/sec. Calculate the service rate (jobs/sec) rate per busy server are constant ( and µ, respectively)
beyond which the system would remain stable. regardless of the state of the system.
21 22

The M/M/s Model

The M/M/1 Model
▪ When the system has just a single server (s = 1),
o the parameters for the birth-and-death process are
o n =  (n = 0, 1, 2, …) and
o µn = µ (n = 1, 2, …).
▪ The resulting rate diagram is shown in Fig. 2.
▪ However, when the system has multiple servers (s > 1),
then according to property 4 of exponential distribution,
o when the mean service rate per busy server is µ,
o and there are n jobs currently in the system
o the overall mean service rate for n busy servers = nµ.
▪ Therefore,
o µn = nµ when n ≤ s,
o whereas µn = sµ when n ≥ s so all s servers are busy.
▪ When the maximum mean service rate sµ exceeds the
mean arrival rate , that is, when

Fig 2: Rate diagram for the M/M/1 model

▪ a queuing system fitting this model will eventually reach
a steady-state condition.
23 24

CSM (CS-438) 6
 The M/M/s Model Results for the Single-Server Case (M/M/1)
▪ For s = 1, the Cn factors for the birth-and-death process
▪ The rate diagram for this case is shown in Fig. 3. reduce to

Fig 3: Rate diagram for the M/M/s model

▪ In this situation, the steady-state results derived earlier

for the general B&D process are directly applicable.
▪ However, these results simplify considerably for this
model and yield closed-form expressions for Pn, L, Lq,
and so forth, as shown next.

25 26

Results for M/M/1 model Results for M/M/1 model

▪ Now find W using little’s formula

27 ▪ Next find Wq using little’s formula 28

CSM (CS-438) 7
 Results for M/M/1 model
▪ When  ≥ µ, so that the mean arrival rate exceeds the
mean service rate, the preceding solution “blows up”
(because the summation for computing P0 diverges).
▪ For this case, the queue would “explode” and grow
without bound.
▪ If the queuing system begins operation with no
customers present,
o the server might succeed in keeping up with
arriving customers over a short period of time,
o but this is impossible in the long run.
29
The M/M/1 Model
PROBLEM
In the head office of a leading commercial bank, there is a
single printer for a fairly large number of continually
arriving print jobs. The system manager has concluded that
print jobs arrive pretty much at random according to Poisson
process. He also has resolved that the time spent in printing a
document follows an exponential distribution.
By projecting the available data for several months, jobs are
estimated to arrive at an average rate of 2 every 30 seconds.
The printer takes an average of 12 seconds to print each job.
a. Would the system attain steady-state? Comment
properly.
b. Find the probability that printer will be idle.
c. If reply of part (a) is positive, figure out the steady-state
results using appropriate queuing model. 31
CSM (CS-438)
The M/M/1 Model
▪ Even when  = µ,
o the expected number of customers in the queuing
system slowly grows without bound over time
because,
o even though a temporary return to no customers
present always is possible,
o the probabilities of huge numbers of customers
present become increasingly significant over time.
30
Queuing Models Involving Nonexponential Distributions
Lecture 30
▪ The assumption of exponential inter-arrival times,
o a reasonable approximation in many situations
o but not when arrivals are carefully regulated.
▪ Furthermore, the service-time distribution frequently
deviates greatly from exponential form, particularly when
the service requirements of the jobs are quite similar.
▪ Therefore, it is important to have other queuing models
that use alternative distributions.
▪ Unfortunately, the mathematical analysis of queuing
models with non-exponential distributions is much more
difficult.
▪ However, it has been possible to obtain some useful
results for a few such models.
32
8
 Queuing Models Involving Nonexponential Distributions Queuing Models Involving Nonexponential Distributions

The M/G/1 Model ▪ Any such queuing system can eventually reach a steady-
state condition if  = / < 1.
Assumptions
▪ The readily available steady-state results for this general
1. The queuing system has a single server and
model are the following:-
2. A Poisson input process (exponential inter-arrival times)
with a fixed mean arrival rate .
3. The jobs have independent service times with the same
probability distribution.
o However, no restrictions are imposed on what this
service-time distribution can be.
▪ In fact, it is only necessary to know (or estimate) the
mean 1/ and variance σ2 of this distribution.
▪ The model does not provide a closed-form expression
for pn because of analytic intractability.
33 34

Models Involving Nonexponential Distributions Models Involving Nonexponential Distributions

▪ Considering the complexity involved in analyzing a model
that permits any service-time distribution, it is remarkable
that such a simple formula is obtained for Lq.
▪ This formula is one of the most important results in queuing
theory because of
o its ease of use and
o the prevalence of M/G/1 queuing systems in practice.
▪ This equation for Lq (or its counterpart for Wq)
o commonly referred to as Pollaczek-Khintchine formula,
o named after two pioneers in development of queuing theory
o who derived the formula independently in the early 1930s.
35
▪ For any fixed expected service time 1/, notice that Lq, L,
Wq, and W all increase as σ2 is increased.
▪ This result indicates that
o the consistency of the server has a major bearing on
the performance of the service facility—
o not just the server’s average speed.
▪ When the service-time distribution is exponential, σ2 =
1/2, and the preceding results will reduce to the the
M/M/1 model.
36

CSM (CS-438) 9
 Uniform Distribution PROBLEM
▪ X is a uniform random variable if the PDF of X is Requests for a page at a web server arrive
1
𝑎 ≤𝑥 <𝑏 randomly according to Poisson distribution with
𝑓𝑋 𝑥 = ቐ(𝑏 − 𝑎)
25 requests per second. The time taken by the
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
server machine to fulfill a request is in between 20
where the two parameters are b > a.
and 50 msecs, uniformly distributed.
Theorem
▪ If X is a uniform random variable with parameters a and b > a.
a) Determine the percentage of time that the web
▪ The CDF of X is
server will be idle?
0 𝑥 ≤𝑎 b) Determine the average backlog for the server.
(𝑥 − 𝑎) c) The average time until a request is fulfilled.
𝐹𝑋 𝑥 = 𝑎<𝑥 ≤𝑏
(𝑏 − 𝑎)
1 𝑥>𝑏
▪ The expected value of X is E[X] = (b+a)/2
▪ The variance of X is Var[X] = (b – a)2/12 37 38

CSM (CS-438) 10