Monitoring and management of SDN is a challenging and active field of research. In this project, we have used Network Tomography techniques to enhance performance of SDN monitoring and management.

1. NETWORK TOMOGRAPHY TO ENHANCE THE
PERFORMANCE OF SOFTWARE DEFINED
NETWORK MONITORING AND MANAGEMENT
K M Sabidur Rahman
with Chaitrali Joshi and Tanjila Ahmed (ECS 273)
9/23/20161

2. 9/23/20162
Agenda
•Network Tomography and SDN
•MiniNet Data Collection
•EM algorithm
•Results from EM implementation in R
•Future work

3. 9/23/20163
Network Tomography
“It is the study of a network’s internal characteristics using information
derived from end point data.”
- Wikipedia
• Statistical theory and algorithms are used to estimate internal
characteristics of the network topology
• Flexibility of SDN can be used to make better results

4. 9/23/20164
Shortcomings of current measurement methods
• Impractical to measure packet delays at every router
• Proprietary routers are inflexible
• Geographic separation of routers
• Storing packet information at every node is expensive
• Memory constraints
• Causes congestion in communication

5. 3/14/165
Experimental Setup
• Topology setup using Mininet
• Single source multiple
receiver
• Controller for SDN
• Python based POX
remote controller
S2
H2 S1
S3
H3 H4
H1

6. 9/23/20166
EM Algorithm
•Iterative method for estimation of parameters in a statistical model
•It estimates the maximum likelihood of parameters
•Where the model depends on unobserved latent variable
•E step: Update the conditional expectation(log likelihood) of the parameter
by the given observation
•M step: Find the parameter that maximizes that quantity

7. 9/23/20167
EM Algorithm
y = observed end-to-end delay data
z = unobserved link delay(latent variable)
p=probability of delays on link i
Complete data likelihood can be factorized as :
where we have to estimate and z from end-to-end delay y

8. 9/23/20168
EM Algorithm
Complete data likelihood is dependent on likelihood of z as conditional pmf of y given z does
not have p parameter.
Where ,number of packets that faces delay j on link i

9. 9/23/20169
EM Algorithm
E Step :
M step:
This is our updated estimate of Pij.

10. 9/23/201610
Results
0
5
10
15
20
25
30
Data Link 1
EM
Numberofpackets
Delay (ms)

11. 9/23/201611
Results
0
5
10
15
20
25
30
Data Link 2
EM Link 2
Numberofpackets
Delay (ms)

12. 9/23/201612
Results
0
5
10
15
20
25
30
Data Link 3
EM
Numberofpackets
Delay (ms)

13. 9/23/201613
Future
•Use more data 100,1000,2000
•Use more complex topology
•Use SDN control mechanism to find the minimum measurement needed for
a network
•Explore temporal effects on delays

14. 9/23/201614