Queueing Theory

Markov Chain
• Markov Process is an important class of
random process which has simple and
highly useful form of dependency among
random variables
• Markov Chain is a special case of Markov
Process where the system can occupy a
finite number of states
• It is a model of the random motion of an
object in a discrete set of possible
locations
• Two versions:
1. Discrete time
Position of the object (state of the
Markov Chain) recorded every unit of time
2. Continuous time
State is observed at all times, t  0
• State of the Markov chain changes
randomly
• The law of future motion of the state
depends only on the present location and
not on previous locations
- Forgets its past locations
• A discrete time Markov chain ( discrete time
discrete state Markov process) is defined as one
which has the property

P[{X(tn+1) = xn+1} / {X(tn) = xn, X(tn-1) =

xn-1, …X(t1) = x1}]

= P[{X(tn+1) = xn+1} / {X(tn) = xn}]

where t1 < t2 < t3 and xi is the ith discrete state

space value
• Prob. that the random variable X takes the
value xn+1 at time step n+1 is entirely
determined by its state value in the
previous time step n and is independent of
its state value in earlier time steps

• Entire past history summarized in its

current state and hence the next state is
determined only by the current state
• We require some specification of the time that
elapses between the transitions
• We have to seek a probability distribution
function for time which would in no way influence
the state transitions
• Only two distribution functions that satisfy this
criterion
Exponential distribution - for continuous time
Geometric distribution - for discrete time
• Markov chain is used to model the progression of
the calls that a telephone network carries & of
packets that a datagram network transports

• Randomness in these models:

1. Time at which users place calls or send packets
2. Length of packets
3. Transmission errors & failures of devices
• Theory of Markov chain tells us how to calculate
the fraction of time that the state of the chain
spends in different locations
• This can be used to estimate
1. The delays and losses of packets in the
network
2. Fraction of the time telephone calls are
blocked
These can be used to design and control the
network
 Discrete time Markov Chain

• Given a set X called “state space”

Elements of X are called “states”
• Set X is countable
• Typical elements of X are i,j,k
• For i  X, there is a list of non negative
numbers, p(i, j), j  X that add up to one
0  p(i, j)  1 for all i, j  X

 p(i, j)  1 for all i  X

jX

Think of P = p(i, j), j  X as a matrix

• This matrix is N x N if X has N elements
• Matrix P is called “ Transition Probability
Matrix”
 Transition Probability Matrix

• Matrix P = p11 p12 p13 … p1N

p21 p22 p23 … p2N
: : : :
pN1 pN2 pN3 … pNN

• Elements in each row add to unity

• Given a probability distribution, 0 on X,
that is a collection of non negative
numbers  0 (i), i  X that add up to 1
( 0 (i) is the probability that during 0th
instant the state is i )
0  0 (i)  1 for all i  X and

 0 (i) = 1
iX
• A random sequence x =  x0 , x1 , x2..  is
defined that takes values in X as follows
P( x0 = i0, x1 = i1,… xn = in ) =
0 (i0). P(i0 , i1). P(i1 , i2). …. P(in-1 , in).
for all n  0 and all i0 , i1,, i2 ,… in  X

 Random sequence x is called a Markov

chain with transition probability matrix P
and initial distribution 0
 State transition diagram
Example 1

State space X =  0,1 

Transition probability P(0 , 1) = a ;
 P(0 , 0) = 1 - a
P(1 , 0) = b ;
 P(1 , 1) = 1 – b
• Transition probability matrix

P= 1–a a
b 1-b
 Example 2

• 4 states (0, 1, 2 & 3 )

State space X =  0,1,2,3 
• If there is an arrow from i to j, P(i,j) > 0
 9

• Path : Succession of arrows such that end

of one is the start of the next arrow

– It is a possible trajectory of the Markov chain

in the state space
 Binary Communication Channel

• A time-invariant binary communication

channel
• xn is the input and xn+1 is the output
• The channel delivers the input symbol to
the output with certain error probability that
depends on the symbol
• Channel error probabilities are α & β
α & β  1/2
• Pxn+1 = 1 | xn = 0 = p01 = α
• Pxn+1 = 0 | xn = 1 = p10 = β
• In a binary symmetric channel, α = β = p
• Probability distribution of xn is calculated as
follows:
1 (i) := P(x1 = i) =  P(x0 = j, x1 = i )
jX

=  0 (j) P( j, i ) ----(1)
jX
• n (i) := P(xn = i) for i  X
• Let n be the row vector with elements
 n (i), i  X
•  1 = [1 (0), 1 (1)]
and 0 = [0 (0), 0 (1)]
• Eqn (1) can be rewritten as
1 = 0 P
• [1 (0), 1 (1)] = [0 (0), 0 (1)] P

= [0 (0), 0 (1)] p(0,0) p(0,1)

p (1,0) p(1,1)

= [0 (0) p(0,0) + 0 (1) p (1,0)

0 (0) p(0,1) + 0 (1) p (1,1) ]
• 1 = 0 P

• 2 = 1 P

• n+1 = n P, n  0

• n = 0 Pn

Pn is the product of P by itself n times

 Balance equation for the transition
matrix
• Probability distribution  is invariant for the
transition probability matrix P if
 = P

 If  is invariant for P and if 0 = , then

n =  for all n  0
• For the transition probability matrix,

P= 1–a a
b 1–b ,

the only solution  = [ (0) ,  (1)] for the

balance equation is
 = b , a
a+b a+b
 Continuous time Markov Process
• It can be viewed as a discrete time Markov process
with an infinitesimally small time unit
• Continuous time Markov Process x(t) can occupy
the states e0, e1, … at time t.
• The status of the process at time t is described by
x(t)
• Suppose the process x(t) is in state ei at time t0
• The probability that the process goes into the
state ej at time t0+t is given by
P{x(t0+t) = ej | x(t0) = ei}
and this probability is independent of the
behavior of the process prior to the instant t0
• If x(t) is a homogeneous Markov process, this
transition probability from state ei to state ej does
not depend upon t0 but depends only on the
elapsed time t between the transitions
• pij(t) = P{x(t0+t) = ej | x(t0) = ei}
• In particular
pij(t) = P{x(t) = ej | x(0) = ei} where
pi(0) = P{x(0) = ei} represents the initial probability
distribution of the states
• For all states ei, ej
0  pij(t)  1

 pij(t) = 1
j
• The unconditional probability of the event
“x(t) is in state ej” is

pj(t) = P{ x(t) = ej }
=  P{ x(t) = ej | x(0) = ei} P{x(0) = ei }
i
=  pi(0) pij(t)
i
• In general, for arbitrary t and s,

pij(t+s) = P{ x(t+s) = ej | x(0) = ei }

=  P{ x(t) = ek | x(0) = ei}. P{ x(t+s) = ej | x(t) = ek }

k

=  pik(t) pkj(s)
k
 Poisson Process
• Serves as a mathematical model for a
wide range of phenomena including the
arrival of calls
• We consider point events occurring singly
in time
•  is a constant (> 0) with dimension time-1
which we call the rate of occurrence
• If N (t, t+∆t) is the number of events in the
interval (t, t+∆t), as ∆t  0,

Prob {N (t, t+∆t) = 1 } =  ∆t + o(∆t),

Prob {N (t, t+∆t) = 0 } = 1 -  ∆t + o(∆t),
Prob {N (t, t+∆t) > 1 } = o(∆t),
where o(∆t) denotes a function tending
to zero more rapidly than ∆t

We suppose that N (t, t+∆t) is independent

of occurrence in (0,t).
• Prob { N(t) = i } = e-t(t)i
i!
• The mean and variance of the Poisson
distribution are equal to t
 Queueing Theory
• Several fundamental models have been
developed for the analysis of “resource
sharing” systems

• Customers demand resources at random

times and use the resources for variable
durations
• When all the resources are in use, arriving units
(customers) form a line or “queue” to wait for
resources to become available

• Queueing theory deals with the analysis of these

systems
 We are interested in the following performance
measures
 Time spent in the system
 Number of customers in the system
 Fraction of arriving customers that are lost or
blocked
 Average number of messages / second that
pass through the system (throughput)
• The source from which the arriving units come
may be finite or infinite

• An arrival may consist of a single unit or in bulk

• The service system may have either a limited or

unlimited capacity for holding units (waiting room
capacity)

• Service may be rendered either singly or in bulk

Arrivals and departures at a queue

ti – arrival instant

si - service times
N – Number of customers in the system
Basic features of a queue:

1. Input process
2. Service mechanism
3. Queue discipline
4. Server‟s capacity
 1. Input process
• Input process specifies the probability law
governing the arrival statistics of the customers
at the server at times t1,t2,… tn
• n = tn+1 - tn is the inter arrival time between
(n+1)th and nth customers
• The input process is specified by the probability
distribution of the sequence of arrival instants
tn and the sequence of inter arrival times n
• Simplest model for the input process:

 Arrival times follow a Poisson process with

parameter 

Inter arrival times n are exponential

random variables
P[A(t) = k] = (t)k e-t for k=0, 1,2,..
k!
 Input process is Markovian or memoryless
 2. Service mechanism

• It is specified by the sequence of service

times sn, where sn is the time required to
serve the nth customer
• Simplest models:
Constant service duration sn =T or
Exponential distribution with parameter 
 3. Queue discipline

• Rule by which the arriving units form a queue

• The manner in which they behave while waiting
(patient or impatient customers)
• Type of service offered at the server
• Usual discipline:
FIFO
 4. Service System
• May have one or several channels that provide
service at the same or different rates
• May have limited or unlimited capacity for
holding waiting units
• In a single channel the traffic intensity specified
by
 =  = mean arrival rate (no. of arrivals / time)
 mean service rate (no. served / time)
 Queueing model

• Basic elements of a queueing system

• Customers arrive at the system with inter arrival
times 1, 2,…..n

• We assume that inter arrival times are

independent random variables with the same
distribution
 Queueing system classification
• Universally used notations to specify queues-
3 part or 4 part symbol
Arrival process / service time / servers / Max occupancy
1st symbol – Input process (inter arrival distribution)
2nd symbol – Service mechanism (service time
distribution)
3rd symbol – No. of channels or servers used
4th symbol – Used only if the system has a limited
holding capacity for waiting items
• Symbols used to specify the input process and
service mechanism:
M : Poisson or exponential (Markovian or
memoryless )
D : Deterministic or regular
G : General (Arbitrary service time distribution
function )
GI : Arbitrary Independent inter arrival
distribution function
 Examples:

• M/G/r:
- Queue with Poisson arrival
- No assumption about service time distribution
- r number of servers
 M/M/1/K
 Queue with Poisson arrival
• Exponential service time distribution
• 1 server
• K packets are allowed in the system
 Arrival rates and traffic load
definitions
• Let A(t) be the number of arrivals at the system
in the interval from time 0 to „t‟

• B(t): Number of blocked customers

• D(t): Number of customer departures

• Number of customers in the system at time t is

N(t) = A(t) – D(t) – B(t)

• Assumption – System was empty at t = 0
• Long term arrival rate

 = lim A(t) customers / second

t t
• Throughput is the long term departure rate

Throughput = lim D(t) customers / second

t t
• The average number in the system is
t
E(N) = lim 1 ∫0 N(t) dt customers
t t

• Fraction of blocked customers is

Pb = lim B(t)
t   A (t)
 Arrival at a system

• The first customer arrives at 1 , A(t) goes from 0 to 1

• Second arrival is 2 seconds later
• nth customer arrival is at time 1 + 2 + …+ n
• i is the time between the arrival of the (i-1)th and
ith customer

• The arrival rate up to the arrival of nth customer

is n / (1 + 2 + …+ n) customers / second

• Long term arrival rate is

 = lim n
n (1 + 2 + …+ n)
= lim 1 = 1
n   (1 + 2 + …+ n)/ n E[]
• We assume that all the inter arrival times have
the same probability distribution and their
average or expected value is E[]

• Thus the average arrival rate is given by the

reciprocal of the average inter arrival time
 Little‟s Formula

• It relates the average time spent in the system

E[ T ] to the arrival rate  and the average
number of customers in the system E[N]

E[N] =  E[ T ]
• Assume that the system does not block any
customers
• A(t) : Cumulative arrivals to the queue at time t

• D(t) : Cumulative departures

• Number of customers waiting in the system at

time t is

L(t) = A(t) – D(t)

• Customers arrive at time tj, j = 1,2,…

• A(t) represents the number of such arrival times

Arrivals and departures for an FIFO queueing system
• At each arrival A(t) increases by 1

• Customers depart at times tj‟  tj

• Service discipline is FIFO

• Departures coincide with the arrivals when the

system is empty
• Wj is the wait time that each customer spends in
the system between arrival and departure
• Consider a starting time 0 and a later time  at
both of which A(t) = D(t)
( Instants t2‟ , t5‟ , t6‟ )
• Number of arrivals in the interval  is
n() = A() – A(0) ----- (1)
• Mean arrival rate in the interval (0, ) is
() = n() /  ----- (2)
• Cumulative cross hatched area in the interval
(0, ) is

∫0 L(t) dt where L(t) = A(t) – D(t)

• This area is made up of a series of rectangles of

unity height and width Wj
n() 
 Wj = ∫0 L(t) dt ---- (3)
j=1
• Average waiting time in the interval (0, ) is

n()
W( ) =  Wj / n() ---- (4)
j=1

• Average number of customers in the system in

the interval (0, ) is


L( ) = ∫0 L(t) dt / ---- (5)
 
• L( ) = ∫0 L(t) dt / 

Using eqn (3) ,

n()
• L( ) =  Wj /
j=1
Using eqn (4),
L( ) = n() W( ) / 
= [n() / ] W( )
= () W( )
• L(  ) =  () W( ) ---- (6)

• This is the Little‟s formula derived for a special

case of FIFO service and for a finite interval (0, )

• Let   and assume that the quantities of

interest approach definite limits:

W( )  W, ()  , L( )  L

• We then get Little‟s formula

L =  W

L is the average number of customers

in the queueing system
 is the arrival rate
W is the average waiting time

This is generally applicable for any service

discipline
 M/M/1 Queue

• Simplest model of a queue

• Poisson arrival
• Exponential service time statistics
• Single server
• Infinite queue length
 M/M/1 Queue

• Arrival process is Poisson with parameter 

• Service time distribution (packet length or call holding
time) is exponential with parameter 
 k  
k  
• Inter-arrival time distribution
p (t )  et
• Service time distribution
t0
p (t  t0 )    e t dt  1  e t0
0
• Let pn be the probability of state at queue
ie. probability that there are n customers
(packets or calls) in the queue including the one
in service
• Once we find pn at the queue, the following
statistical properties can be determined
 Average queue occupancy
 Probability of blocking for a finite queue
(not applicable for M/M/1 queue)
 Average throughput
 M/M/1 state time diagram

• The probability pn(t+∆t) that there are n customers in the

system at time (t+ ∆t) may be found in terms of the
corresponding probabilities at time t
• If the queue is in state n at time t+ ∆t, it could
have been in states n-1, n or n+1 at time t.
(Assume n  1)

• pn(t+∆t) is the sum of the mutually exclusive

probabilities that the queue was in states n-1, n
or n+1 at time t, multiplied by the probability of
arriving at state n at t+ ∆t
• If the inter arrival times are exponential,
P[ 1 arrival in ∆t ]=  ∆t + o∆t

• Similarly
P[ 0 arrival in ∆t ]= 1-  ∆t + o∆t
• Since the service times also have an
exponential distribution,
P[ 1 departure in ∆t ]=  ∆t + o∆t

• Similarly
P[ 0 departure in ∆t ]= 1- ∆t + o∆t
• pn(t+∆t) = pn(t) [(1-  ∆t )(1- ∆t) + o∆t]
+ pn-1(t) [  ∆t (1- ∆t) + o∆t]
+ pn+1(t) [  ∆t (1-  ∆t ) + o∆t]
------ (1)
• Simplifying and dropping o∆t terms,

pn(t+∆t) = [1- ( +  )∆t ] pn(t) +  ∆t pn-1(t)

+  ∆t pn+1(t) ------- (2)

This equation can be used to study time

dependent (transient) behavior of the M/M/1
queue
 State diagram of M/M/1 queue

• Transitions to the adjacent states only

• There is a rate  of moving up one state due to
an arrival and a rate  of moving down due to
service completion
•  ∆t is the prob. of moving up one state and
 ∆t is the prob. of moving down one state
 Balance equation
• (  +  ) Pn =  Pn-1 +  Pn+1 n1
• LHS represents rate of leaving state n, given the
system was in state n with probability Pn
• RHS represents the rate of entering state n,
from either state n-1 or state n+1
• For stationary state probabilities to exist, the two
rates must be equal
State transition diagram of M/M/1
• If the system is stable, long term transition rate

from n to n+1 must be equal to the long term

transition rate from n+1 to n

pn+1  ∆t = pn  ∆t --- (5)

• Then pn+1 = ( / ) pn n = 0,1,2,…

• pn = ( / )n p0 = n p0 n = 0,1,2,…
--- (6)
• To find p0 we use the fact that the probabilities
must add up to 1
1 = p0 + p1+ p2 + ……
= p0 ( 1 +  + 2 + 3 +…. )
= p0 1 for  < 1 --- (7)
1–
Hence
p0 = 1– --- (8)
• Probabilities for the number of customers in the
system
P[N(t) = n] = pn
• From eqn (6)
p n = n p 0
• Substituting eqn (8) in (6),

pn = (1 – ) n
• pn = (1 – ) n is the equilibrium state
probability solution for the M/M/1 queue
• This probability distribution is called geometric
distribution
•  < 1 is the necessary condition
• Its significance is that for equilibrium to exist,
the arrival rate or load on the queue must be
less than the capacity 
Example – M/M/1 state probabilities with  = 0.5
• p0 = 1 – 

 This is the probability that the queue is empty

 Hence the probability that the queue is non

empty is  which is simply the utilization
Queueing system variables
Average no. of customers in the M/M/1
system
• The average number of customers in the system
E[N] is given by

E[N] =  n pn
n=0

E[N] =  n (1 – ) n
n=0

=  = 
(1 – ) ( - )
 Average delay in the M/M/1 system

• By Little‟s formula

L =  W

L is the average number of customers

in the queueing system
 is the arrival rate
W is the average waiting time
• Average delay E[T] is given by

E[T] = E[N]


= / = 1/
(1 – ) (1 – )
 Average waiting time
• Average “waiting time” is obtained as
E[W] = E[T] – E[X]
E[T] is the average delay in the system
E[X] is the average service time (1 /  )

= 1/ _ 1/
(1 – )

= (1 / ) 
(1 – )
M/M/m/m queue (Erlang‟s model)
• Poisson Arrival Process
• Exponential Service time distribution
• „m‟ number of servers
• „m‟ number of subscribers in the system
• No buffer space.
• Erlang had originally used this loss model to
investigate the distribution of busy channels in
telephone system

• Up to m incoming calls are handled at once

• An incoming call goes through if at least one

server is free, otherwise it is rejected
• Each trunk is viewed as a server and the
connection time is viewed as the service time X
Service rate  = 1 / E[X]
• The state of this system is given by N(t), the
number of trunks in use
• Each connection request increases N(t) by 1 and
each connection release decreases N(t) by 1
• State of the system can take on the values 0, 1,
2, ….., m.
• The probability of a connection request in the
next ∆t seconds is given by  ∆t
• M/M/m/m system differs from the M/M/1 system
in terms of the departure rate
If N(t) = n, n servers are busy, each server will
complete its service in the next ∆t seconds with
probability .∆t
• The probability that one of the connections
completes its service and that the other n-1
continue their service in the next ∆t seconds is
given by
n( ∆t ) (1 -  ∆t )n-1  n  ∆t
• The probability that two connections will complete
their service is proportional to ( ∆t )2 which is
negligible relative to ∆t
• Therefore the departure rate when N(t) = n is n
• j = jm
=0 j> m

 j = j. j<m
= m. j=m
State transition diagram for M/M/m/m

• Equating the transition rates from n+1 to n and n

to n+1 for a stable system,
pn+1 (n+1)  ∆t = pn  ∆t --- (1)
• pn+1 =  pn n=0,1,2,…,m-1
(n+1) 

• Repeated applications of the recursion gives

pn+1 = ( / )n+1 p0 n=0,1,2,…,m-1

(n+1)!
• 1 = p0 + p1+ p2 + ……+ pm

m n
= p0 
n=0 n!

n / n! for n=0,1,2,…,m
 pn =
m k

k=0 k!

The probability of blocking is given by pb = pm which

leads to Erlang B formula
 M/M/ (Infinite server)

• Number of servers is assumed to be unlimited

so that there is no queue, and an arriving item is
instantly served
• j =  and j = j  for j = 0,1,2,…. so
that steady state solutions can be obtained by
letting m   in the M/M/m/m system
 1
p0 =  (/)n = e- (/)

n=0 n!

And pn = p0 (/)n = e- (/) (/)n

n! n!
n= 0,1,2,….

Thus in steady state, the number of items in an

M/M/  queue is Poisson distributed with
parameter /
 State-dependent Queues
• Number of packets (or customers) in the
system is n.

• Probabilities of a new packet (customer)

arriving or leaving in time Δt, is λn Δt and
μn Δt similar to Poisson, but depends on n.

• Still called Markov processes

 0 1 2 n-1 n n+1

… …
0 1 2 n-1 n n+1
1 2 3 n n+1 n+2

• Average flow between states now

depends on n
• The probabilities do not depend on what
happened in the past: this is still a
memoryless system.
• The average flow between two states
must equal zero
λnpn = μn+1pn+1

• If we start with n = 0, we get

p1 = p0λ0/ μ1


n 1
i
• And in general pn  p0 i 0

 
n
i 1 i
• The longer the queue, the less likely a new
packet will arrive
• Assume λn = λ/(n+1) for n = 0, 1, 2, 3,…
• λ is some constant
• The service process is still a simple
Poisson process with service rate, μ
• We can calculate pn by

n 1
 n n
pn  p0 i 0 i
 p0 n  p0
   n!
n
n!
i 1 i

where ρ = λ/μ
• Sum of state probabilities is 1.
 
 n

p
n 0
n  p0 
n 0 n!

 p0e  1

• Therefore p0=e-ρ
• Average queue length is
 
n 
 n1
E (n)   npn   np0  e 


n 0 n 0 n! n 1 (n  1)!

• This is always finite, no matter how

large λ is
• This is different from the result for the
Poisson process which can go to ∞
• The average arrival rate has to be
found from all states
 
p0  n
   n p n  
n 0 n  0 ( n  1) n!

 e    n 
 n 1
  e 
 (n  1)!
n 0 (n  1) n! n 0
 
 e (e  1)
  (1  e   )
• We see that the average arrival rate, μ(1 -
e-ρ) is always less than the average
service rate, μ.

• This explains why the queue never goes

unstable, as it can for Poisson arrivals

• For high ρ, γ μ
• We find the average delay time from
Little‟s formula
• E(T) = E(N)/γ = ρ/(μ(1- e-ρ))
 M/G/1 queue
• Poisson arrival process
• Customer service times have a general
distribution- not necessarily exponential as in the
M/M/1 system
• Single server
• Infinite queue length

μ=1/E[X]
• Examples:
1. Constant interarrival - M/D/1
2. Erlang ,, - M/Er/1
3. Exponential ,, - M/M/1
4. Hyperexponential ,, - M/H/1
• Let Xi be the service time of the ith arrival.
• We assume that random variables (X1,
X2,…) are identically distributed, mutually
independent, and independent of the inter-
arrival times.
• Let
1
X  E{ X }   Average service time

X 2  E{ X 2 }  second moment of service time
• Pollaczek-Khinchin (P-K) formula
X2
W
2(1   )
W is the expected customer
waiting time in queue and

  X

• Given the P-K formula, the total waiting
time in queue and in service is
X2
TX
2(1   )
• Applying Little‟s formula, the expected
number of customers in the system, N is

2 X 2
N 
2(1   )
 Generalization in M/G/1 model
1. Priority queueing systems
M/G/1 can be generalized to the case
where customers can belong to one of K
priority class.
When a customer arrives at the
system, it joins the queue of its priority class.
2. M/G/1 model with vacation

– Server goes on vacation (becomes

unavailable) whenever it empties the queue.
– If upon returning from vacation, the server
finds that the system is still empty, the server
takes another vacation.
 Birth – death process

• Restriction applied to the Markov chain that

transitions can occur only to the adjacent
states
• The number in the population in a birth-death
process is a random variable and represents
the state value
• B – D process is useful in analysis of
telecommunication network
State transitions of a B – D process
• Telecom network can be modelled as a B-D
process where

No. of servers  population

Call request  birth
Call termination  death
• In order to analyze a B–D process, choose a
time interval t small enough such that

1. There can be only one state transition in that

interval
2. Only one arrival or termination, but not both
3. There may be no arrival or termination
• Assumptions:
1. Prob. of an arrival or termination in a particular
interval is independent of what has happened in the
earlier time interval
2. Prob. of an arrival is directly proportional to the
time interval

• Dynamics of the switching system:

• Pk(t) = prob. that the system is in state k at time t
( k servers are busy at time t)
• k = call arrival rate in state k
• k = call termination rate in state k
• In interval t,
P[exactly one arrival ] =  t
P[exactly one termination ] =  t
P[ no arrival] =1 -  t
P[ no termination ] = 1 -  t

• Prob of finding the system in state k at time t+ t is

Pk(t+ t ) = Pk-1(t) k-1 t + Pk+1(t)  k+1 t +
(1 - k t ) (1 -  k t ) Pk(t) ----- (1)
Expanding this and ignoring the (t )2 term being small,
Pk(t+ t ) = Pk-1(t) k-1 t + Pk+1(t)  k+1 t +
[1 - k t -  k t ] Pk(t)
Pk(t+ t ) = Pk-1(t) k-1 t + Pk+1(t)  k+1 t
- ( k +  k ) Pk(t) t + Pk(t)

• Dynamics of the system => rate of change of Pk with time

Pk(t+ t ) - Pk(t) = Pk-1(t) k-1 + Pk+1(t)  k+1 - ( k +  k ) Pk(t)
t

• If t  0,
d Pk(t) = Pk-1(t) k-1 + Pk+1(t)  k+1 - ( k +  k ) Pk(t)
dt ----- (2)
This is the differential equation governing dynamics of a B – D
process