Scribd
Upload
17 views

Queueing Systems: Lecture 5

This document provides a summary of a lecture on queueing systems. It discusses: 1) A fundamental result for analyzing open acyclic networks of M/M/m queues with infinite queue capacities and Poisson arrivals. 2) Examples of using state transition diagrams to analyze Markovian queueing systems and networks, including two M/M/1 queues with finite capacity and an M/Ek/1 system. 3) Approaches for analyzing dynamic queueing systems, including simulation and numerical solutions of equations describing the evolution of queues over time.

Uploaded by

Arunsingh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
17 views

Queueing Systems: Lecture 5

This document provides a summary of a lecture on queueing systems. It discusses: 1) A fundamental result for analyzing open acyclic networks of M/M/m queues with infinite queue capacities and Poisson arrivals. 2) Examples of using state transition diagrams to analyze Markovian queueing systems and networks, including two M/M/1 queues with finite capacity and an M/Ek/1 system. 3) Approaches for analyzing dynamic queueing systems, including simulation and numerical solutions of equations describing the evolution of queues over time.

Uploaded by

Arunsingh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

Queueing Systems: Lecture 5

Amedeo R. Odoni

October 30, 2006

Lecture Outline
• A fundamental result for queueing networks
• State transition diagrams for Markovian
queueing systems and networks: examples
• Examples
• Dynamic queueing systems and viable
approaches
• Qualitative discussion of behavior

Reference: Sections 4.10, 4.11 + material in


handout
A result which is important in analyses

of queueing networks

Let the arrival process at a M/M/m queueing


system with infinite queue capacity have
parameter λ. Then, under steady state
conditions (λ<mμ) the departure process from
the queueing system is also Poisson with
parameter λ.
Implication: greatly facilitates analysis of open
acyclic networks consisting of M/M/m queues
with infinite queue capacities.
The bad news: result holds only under exact set
of conditions described above.

Open acyclic network of M/M/.


systems

#2: μ=2
λ=4/3 1 negative
exponential
server

P=1/3
λ=4 #1: μ=3 (per λ=4/3
server)
M/M/2
Q=2/3
#3: μ=6 λ=4
+ 1 negative
λ=8/3 exponential
server
State transition diagrams for
queuing systems and networks
• When external arrivals are Poisson and
service times are negative exponential, many
complex queueing systems and open acyclic
queueing networks can be analyzed, even
under dynamic conditions, through a
judicious choice of state representation.
• This involves writing and solving (often
numerically) the steady-state balance
equations or the Chapman-Kolmogorov first-
order differential equations.
• The “hypercube model” (Chapter 5 of Larson
and Odoni) is a good example.

Example 1: Two M/M/1 Queuing Systems,


Each with Finite Queue Capacity

Poisson
arrivals; λ One server; One server;
neg. exp’l; μ 1 neg. exp’l; μ 2

No queuing No queuing
space space

Note: The queuing system on the right may “block” the one on
the left.
Example 2: M/Ek/1 System, with
system capacity for total of N users

See distributed notes.

Example 3: Two Types of Users


and Non-Preemptive Priorities

Type 1 customers;
Poisson arrivals;
Neg’ve exp’l service;
rate λ1 μ1

μ2
Type 2 customers;
Poisson arrivals;
rate λ2

Type 1 customers have


non-preemptive priority

See distributed notes.


Comparison of August Weekday Peaking
Patterns
1993 vs. 1998 (3 Hour Average)
Operations
130

120

110
1993 1998
100

90
80

70
60

50

40

30

20

10

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Hour

Two common “approximations” (??)


for dynamic demand profiles
1. Find the average demand per unit of time
for the time interval of interest and then
use steady-state expressions to compute
estimates of the queuing statistics.
[Problems?]
2. Subdivide the time interval of interest into
periods during which demand stays
roughly constant; apply steady-state
expressions to each period separately.
[Problems?]
Problems

with the Approximate Methods

• Problems with Approach 1:


1. For cases in which demand varies significantly (e.g.,
>10% from overall average value) the delay estimates
can be VERY poor
2. Will underestimate overall average delay, possibly by a
lot
• Problems with Approach 2:
1. May not have ρ < 1, for some intervals; then what?
2. Time to reach “steady state” is large for values of ρ
which are close to 1; therefore “steady state”
expressions may be very poor approximations when
intervals are relatively short
3. Approach does not take into consideration the
“dynamics” of the demand profile

The Two Viable Approaches

1. Simulation:
• High level of detail
• May be only viable alternative for complex
systems
• Statistical significance of results?
2. Numerical solution of equations
describing the evolution of queueing
system over time:
• Increasingly practical
• May provide lots of information, such as Pn(t)
Dynamic Behavior of Queues

1. The dynamic behavior of a queue can be complex and


difficult to predict
2. Expected delay changes non-linearly with changes in
the demand rate or the capacity
3. The closer the demand rate is to capacity, the more
sensitive expected delay becomes to changes in the
demand rate or the capacity
4. The time when peaks in expected delay occur may lag
behind the time when demand peaks
5. The expected delay at any given time depends on the
“history” of the queue prior to that time
6. The variance (variability) of delay also increases when
the demand rate is close to capacity

The dynamic behavior of a queue; expected delay


for four different levels of capacity

Demand
Delays (mins) (movements)
40 120
35 105
30 90
25 75

20 60

15 45

10 30

5 15
0
0

0
00

00

00

00

00

:0

:0

:0

:0

:0

:0

:0
1:

3:

5:

7:

9:

11

13

15

17

19

21

23

Dem R1 R2 R3 R4

(R1= capacity is 80 movements per hour; R2 = 90; R3 = 100; R4 = 110)


Two Recent References on Numerical

Methods for Dynamic Queueing Systems

• Escobar, M., A. R. Odoni and E. Roth, “Approximate


Solutions for Multi-Server Queueing Systems with
Erlangian Service Times”, with M. Escobar and E. Roth,
Computers and Operations Research, 29, pp. 1353-1374,
2002.
• Ingolfsson, A., E. Akhmetshina, S. Budge, Y. Li and X.
Wu, “A Survey and Experimental Comparison of Service
Level Approximation Methods for Non-Stationary M/M/s
Queueing Systems”, Working Paper, U. of Alberta.
http://www.bus.ualberta.ca/aingolfsson/working_papers.htm

You might also like