Scribd
Upload
57 views11 pages

Queueing Systems: Lecture 3: Announcements

The document summarizes a lecture on queueing systems with Poisson arrivals and general service times (M/G/1 queues). Key points include: - M/G/1 queues can be modeled as discrete-state, continuous-time Markov processes using the concept of epochs. - Transition probabilities between states at epochs can be derived in terms of arrival distributions. - Expected values like average queue length, waiting time, and utilization can be calculated in closed form. - Variability in service times increases expected delays. - An example runway queue is analyzed to estimate performance. - Non-preemptive priority queues with multiple classes are introduced.

Uploaded by

Arunsingh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
57 views11 pages

Queueing Systems: Lecture 3: Announcements

The document summarizes a lecture on queueing systems with Poisson arrivals and general service times (M/G/1 queues). Key points include: - M/G/1 queues can be modeled as discrete-state, continuous-time Markov processes using the concept of epochs. - Transition probabilities between states at epochs can be derived in terms of arrival distributions. - Expected values like average queue length, waiting time, and utilization can be calculated in closed form. - Variability in service times increases expected delays. - An example runway queue is analyzed to estimate performance. - Non-preemptive priority queues with multiple classes are introduced.

Uploaded by

Arunsingh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 11

Queueing Systems: Lecture 3

Amedeo R. Odoni

October 18, 2006

Announcements

• PS #3 due tomorrow by 3 PM
• Office hours – Odoni: Wed, 10/18, 2:30-4:30;
next week: Tue, 10/24
• Quiz #1: October 25, open book, in class;
options: 10-12 or 10:30-12:30
• Old quiz problems and solutions: posted Thu
evening along with PS #3 solutions
• Quiz coverage for Chapter 4: Sections 4.0 –
4.6 (inclusive); Prof. Barnett’s lecture NOT
included
Lecture Outline
• Remarks on Markovian queues
• M/E2/1 example
• M/G/1: introduction, epochs and transition
probabilities
• M/G/1: derivation of important expected values
• Numerical example
• Introduction to M/G/1 systems with priorities

Reference: Section 4.7

Variations and extensions of


birth-and-death queueing systems
• Huge number of extensions on the previous
models
• Most common is arrival rates and service
rates that depend on state of the system;
some lead to closed-form expressions
• Systems which are not birth-and-death, but
can be modeled by continuous time, discrete
state Markov processes can also be
analyzed [“phase systems”]
• State representation is the key (e.g. M/Ek/1 or
more than one state variables – P.S. #3)
M/G/1: Background

• Poisson arrivals; rate λ


• General service times, S; fS(s); E[S]=1/μ; σS
• Infinite queue capacity
• The system is NOT a continuous time Markov process
(most of the time “it has memory”)
• We can, however, identify certain instants of time
(“epochs”) at which all we need to know is the number
of customers in the system to determine the probability
that at the next epoch there will be 0, 1, 2, …, n
customers in the system
• Epochs = instants immediately following the completion
of a service

M/G/1: Transition probabilities for


system states at epochs (1)
N = number of customers in the system at a random
epoch, i.e., just after a service has been completed
N' = number of customers in the system at the
immediately following epoch
R = number of new customers arriving during the
service time of the first customer to be served after
an epoch

N' = N + R – 1 if N > 0

N' = R if N = 0
• Note: make sure to understand how R is defined
Epochs and the value of R

N Between t1 and t2, R=2


Between t5 and t6, R=0

t
t1 t2 t3 t4 t5 t6

M/G/1: Transition probabilities for


system states at epochs (2)
• Transition probabilities can be written in
terms of the probabilities:

P[number of new arrivals during a service


time = r] =
r − λt
∞ (λt) ⋅ e
α r = ∫0 ⋅ f S (t) ⋅ dt for r = 0, 1, 2,...
r!
• Can now draw a state transition diagram at
epochs
A Critical Observation

• The probabilities P[N=n] of having n customers


in the system at a random epoch are equal to the
steady state probabilities, Pn, of having n
customers in the system at any random time!
• The PASTA property: “Poisson arrivals see time
averages”
• Can use simple arguments to obtain (as for
M/M/1 systems):

P0 = 1- (λ / μ) = 1- ρ and E[B] = 1/(μ – λ) = 1/(1- ρ)

• Can also derive closed-form expressions for L, W,


Lq and Wq

Probability the Server is Busy

• Suppose we have been watching the system for a long


time, T.
ρ, the utilization ratio, is the long-run fraction of
time (= the probability) the server is busy; this means,
assuming the system reaches steady state:

amount of time server is busy λ ⋅T ⋅ E[S] λ


ρ= = = λ ⋅ E[S] =
T T μ
Idle and Busy Periods; E[B]

Observe a large number, N, of busy periods:


No. of
N ⋅ E[ B] E[ B]
customers ρ= =
(N ⋅ E[ B] + N / λ ) ( E[ B] + 1 / λ )
ρ /λ 1
E[B ] = =
1− ρ μ − λ

t
I B I B I B

Derivation of L and W: M/G/1


(assumes FCFS)

T = total amount of time a randomly arriving


customer j will spend in the M/G/1 system
T1 = remaining service time of customer
currently in service
T2 = the time required to serve the customers
waiting ahead of j in the queue
T3 = the service time of j

• Clearly:

T = T1 + T2 + T3

W = E[T] = E[T1] + E[T2] + E[T3]

Derivation of L and W: M/G/1 [2]

• E[T3] = E[S]
• Given that there are already n customers in the system
when j arrives (and since one customer is being served
while n–1 are waiting)
E[T2 | n] = (n −1) ⋅ E[S] , n ≥ 1

E[T2 | n] = 0, n = 0

• Thus,
⎡ ⎤
E[T2 ] = ∑ E[T2 | n] ⋅ Pn = ∑ (n −1) ⋅ E[S] ⋅ Pn = E[S] ⋅⎢ ∑ nPn − ∑ Pn ⎥
n n≥1 ⎢⎣n≥1 n≥1 ⎥⎦

E[T2 ] = E[S] ⋅ L − E[S] ⋅ ρ

Derivation of L and W: M/G/1 [3]

• From our “random incidence” result (2.66):

σ S2 + ( E[ S ]) 2
E[T1
| n] = , n ≥1
2 ⋅ E[ S ]
E[T1 | n] = 0, n = 0

• Thus, giving:

σ S2 + ( E[ S ]) 2 σ S2 + ( E[ S ]) 2
E[T1 ] = ∑ E[T1 | n] ⋅ Pn = ∑ ⋅ Pn = ⋅ρ
n n≥1 2 ⋅ E[ S ] 2 ⋅ E[ S ]
Derivation of L and W: M/G/1 [4]

• So we finally have:

L = λW (Little’s theorem) (1)

W = E[T] = E[T1] + E[T2] + E[T3] (2)

and solving (1) and (2), we obtain:

ρ 2 + λ 2 ⋅ σ S2
L=ρ+ ( ρ < 1)
2(1 − ρ )

ρ 2 + λ 2 ⋅ σ S2
1
W = +
μ 2λ (1 − ρ )

Expected values for M/G/1

ρ 2 + λ 2 ⋅ σ S2
L=ρ+ ( ρ < 1)
2(1 − ρ )

1 ρ 2 + λ2 ⋅ σ S2

W = +
μ 2λ (1 − ρ )

ρ 2 + λ2 ⋅ σ S2 ρ 2 (1 + C S2 ) 1 ρ (1 + C S2 )
Wq = = = ⋅ ⋅
2λ (1 − ρ ) 2λ (1 − ρ ) μ (1 − ρ ) 2

ρ 2 + λ2 ⋅ σ S2 σS
Lq = Note : C S = = μ ⋅σ S
2(1 − ρ ) E[ S]
Dependence on Variability (Variance)

of Service Times

Expected delay

Demand

ρ = 1.0

Runway Example
• Single runway, mixed operations
• E[S] = 75 seconds; σS = 25 seconds
μ = 3600 / 75 = 48 per hour
• Assume demand is relatively constant for
a sufficiently long period of time to have
approximately steady-state conditions
• Assume Poisson process is reasonable
approximation for instants when demands
occur
Estimated expected queue length
and expected waiting time

λ (per hour) ρ Lq Lq Wq Wq
(% change) (seconds) (% change)
30 0.625 0.58 69
30.3 0.63125 0.60 3.4% 71 2.9%

36 0.75 1.25 125


36.36 0.7575 1.31 4.8% 130 4%

42 0.875 3.40 292


42.42 0.88375 3.73 9.7% 317 8.6%

45 0.9375 7.81 625


45.45 0.946875 9.38 20.1% 743 18.9%

Can also estimate PHCAP ≅ 40.9 per hour

M/G/1 system with non-preemptive


priorities: background
• r classes of customers; class 1 is highest
priority, class r is lowest
• Poisson arrivals for each class k; rate λk
• General service times, Sk , for each class;
fSk(s); E[Sk]=1/μk; E[Sk2]
• FIFO service for each class
• Infinite queue capacity for each class
• Define: ρk = λk /μk
• Assume for now that: ρ = ρ1 + ρ2 +….+ ρr <1
A queueing system with r priority
classes
λ1 x x 1

x x x 2
λ2

x k-1
λk−1
Service

λk k Facility

xxx x k+1
λk+1

λr xx r

You might also like