Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

by: Janni Boydon

QUEUEING NETWORK
SIMULATION
 This guide is not originally mine. Please don’t
share outside. ☺

 Recall: Queueing Networks


An Introduction to
JMT Simulator

Queueing Theory Fall, 2020


GICE, National Taiwan University

4
Example: Queueing network (1/3)

• Construct a queueing network model as Fig. 17.


• Double click on the
and see the window as Fig. 18.
• Select routing strategy for Queue 1.

Fig. Fig.
17 18

5
Example: Queueing network (2/3)

• Add performance indices for each queue.

6
Example: Queueing network (3/3)

The simulation results are shown as below.

7
Summary

• The procedure to run a simulation on JMT can be summarized as


follows:
1. Build up the system
2. Define customer class
3. Define system parameters
4. Define performance indices
5. Set simulation parameters
6. Start the simulation
• To learn more details about the JMT simulator, read the manual.

8
• Download JMT
• http://jmt.sourceforge.net/Download.html
• Download Java
• http://www.oracle.com/technetwork/java/javase/downloads/index.html
• JMT user manual
• http://jmt.sourceforge.net/Papers/JMT_users_Manual.pdf
• Documents
• http://jmt.sourceforge.net/Documentation.html
• JSIMgraph demo
• https://www.youtube.com/watch?v=HTtQQVrXpx0&list=PL5yYxD_4CokS18VQIAA
WhzOySQYnPlV2b

9
IE 27 & 28 REVIEW
Example: Two independent samples of two brands of broadband providers
were tested related to downloading time. Which is faster?
Sample X Y
Do we simply get the mean and compare?
1 8.94 9.06
2 11.66 18.80 𝑋ത = 12.60444
3 17.76 15.66 𝑌ത = 17.07100
4 11.27 22.08
5 11.99 12.18 Do we then conclude that X is faster than Y?
6 15.85 18.78
7 16.78 23.51
8 9.75 15.10
9 9.88 22.34
10 8.50 13.20
IE 27 & 28 REVIEW
No! We use statistical hypothesis tests to conclude.
Recall:
Z test
T test w/ unequal variances
T test w/ pooled variance
Paired T test
F test
IE 27 & 28 REVIEW
Example: Two independent samples of two brands of broadband providers
were tested related to downloading time. Which is faster?
Sample X Y First: do a test on variance (F Test).
1 8.94 9.06
2 11.66 18.80
3 17.76 15.66
4 11.27 22.08
5 11.99 12.18
6 15.85 18.78
7 16.78 23.51 Remember how to conclude using p-value?
8 9.75 15.10
Conclusion: the variances are equal.
9 9.88 22.34
10 8.50 13.20
IE 27 & 28 REVIEW
Example: Two independent samples of two brands of broadband providers
were tested related to downloading time. Which is faster?
Sample X Y Second: do the appropriate test on mean.
1 8.94 9.06
2 11.66 18.80
3 17.76 15.66
4 11.27 22.08
5 11.99 12.18
6 15.85 18.78
7 16.78 23.51 Remember how to conclude using confidence interval?
8 9.75 15.10 Remember how to conclude using p-value?
9 9.88 22.34 Conclusion: the means are not equal.
10 8.50 13.20 Conclusion: X is faster than Y.
COMPARING SIMULATIONS
It is basically the same when comparing alternatives of simulations with replicates.
Replicate Current Proposed Say, Current is the Lq when you implemented an
M/M/2 queuing system and Proposed is the Lq when
1 8.94 9.06
you implemented 2 M/M/1 queuing system.
2 11.66 18.80
3 17.76 15.66 Which is better?

4 11.27 22.08 To solve:


5 11.99 12.18 1. Perform F test to determine equality of variances
6 15.85 18.78 2. Perform the appropriate test on means
7 16.78 23.51
8 9.75 15.10 In practical applications, usually confidence
intervals are used to statistically compare.
9 9.88 22.34
If zero is not in the confidence interval on
10 8.50 13.20
difference of means, then the means are
statistically different.
COMPARING SIMULATIONS
General formula of confidence intervals on difference in means.

Sp = pooled variance
Si = stdev of sample i
SD = stdev of
difference of pairs
R = # of replicates
COMPARING MULTIPLE ALTERNATIVES
For multiple alternatives (e.g. 5-10+ alternatives), we resort to ANOVA (to help
identify causality of variation) or Metamodels/Regression (to approximate a
mathematical relationship among factors and responses) otherwise use
pairwise tests.
Example (X-axis: # of servers vs. Y-axis: total waiting + service cost):
COMPARISON OF ALTERNATIVES

“Are the differences between samples big enough to have real meaning
or are they only due to random variation?”
QUEUEING NETWORK
SIMULATION

END

You might also like