OpenFlow enabled networks split and separate the data and control planes of traditional networks. This design commodifies network switches and enables centralized control of the network. Control decisions are made by an OpenFlow controller, and locally cached by switches, as directed by controllers. Since controllers are not necessarily co-located with switches that can significantly impact the forwarding delay incurred by packets in switches. Only very few studies have been conducted to evaluate the performance of OpenFlow in terms of end-to-end delay. In this work we develop a stochastic model for the end to end delay in OpenFlow switches based on measurements made in Internetscale experiments performed on three different platforms, i.e. Mininet, the GENI testbed and the OF@TEIN testbed.

1. Analytical Modeling of End-to-End Delay
in OpenFlow Enabled Networks
Presentor:
Azeem Iqbal
School of Electrical Engineering and Computer Science (SEECS)
National University of Sciences and Technology (NUST)
Applied Network & Data Science Research (AN-DASH) Group
1

2. Agenda
1. SDN Background
2. Problem Statement
3. Motivation
4. Experimental Approach
5. Experimental Setup
6. Results
2

3. 1. SDN Background
2. Problem Statement
3. Motivation
4. Experimental Approach
5. Experimental Setup
6. Results
3

4. What is a data plane?
Data plane (DP):
Packet forwarding
Forward, filter, buffer, mark,
rate-limit, and measure packets
4

5. Data plane (DP):
Packet forwarding
Forward, filter, buffer, mark,
rate-limit, and measure packets
What is a control plane?
Control plane (CP):
Distributed
algorithms
Track topology
changes, compute
routes, install
forwarding rules
5

6. Data plane (DP):
Packet forwarding
Forward, filter, buffer, mark,
rate-limit, and measure packets
Track topology
changes, compute
routes, install
forwarding rules
Control plane (CP):
Distributed
algorithms
Management plane:
Human time scale
Collect measurements and
configure the equipment
What is a management plane?
6

7. 7
Traditional Networking
Data
Plane
Data
Plane
Data
Plane
Data
Plane
Data
Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Data
Plane
• Very fast, e.g., 10+ Gbps
• Implemented in h/w
• Mostly table lookups,
e.g., dest addr == 10 
send out port 7
Per-switch
Control Plane
• Implemented in s/w on
commodity chips (x86)
• Much slower,
• Programs h/w tables
• One copy per device

8. Commodity (x86) Server
Data
Plane
Software-Defined Networking
8
SDN
Application
SDN
Application
SDN
Application
…
Data
Plane
Data
Plane
Data
Plane
Data
Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Per-switch
Control Plane
Software-defined Network (SDN) Controller
1. http://www.networkcomputing.com/networking/inside-
googles-software-defined-network/a/d-id/1234201

9. Commodity (x86) Server
Data
Plane
Software-Defined Networking
9
…
Data
Plane
Data
Plane
Data
Plane
Data
Plane
Software-defined Network (SDN) Controller
Network
Virt.
Monitoring/
Visibility
Traffic Eng.
e.g. Google1
1. http://www.networkcomputing.com/networking/inside-
googles-software-defined-network/a/d-id/1234201

10. Modern switches:
◦ Control plane populates forwarding
tables
◦ Data plane acts based on table
entries
◦ Both run locally on the switch
SDN
◦ Decouple control plane from the
data plane
◦ Data plane on the switch
◦ Control plane elsewhere
(typically separate controller)
◦ Example: OpenFlow
Software Defined Networks
Migrate the Control Plane to a Separate Controller
Switch Chip
dst port
0E 5
dst port
0E 5
0A 1
dst port
0E 5
0A 1
0C 3
Control
Plane CPU
Ports, 1-6
SDN
Controller
This gets smaller,
turns into
controller to
switch chip
translator
Most features
go here
0A->0E0A->0E0A->0C
Table miss,
send to
controller
Install table
entry, send
packet
0C->p3
10

11. Software Defined Networks
What’s the big deal?
Potential Benefits:
◦ Enables innovation
◦ Exploit global network view
◦ Traffic engineering
◦ Traffic steering
◦ Security enforcement
◦ Simpler switches
◦ Co-manage virtual compute, storage, and
network
11

12. OpenFlow
Switch
Data Path (Hardware)
OpenFlow
Any Host
OpenFlow Controller
OpenFlow Protocol (SSL/TCP)
The controller is responsible for
populating forwarding table of the
switch
In a table miss the switch asks the
controller
12

13. OpenFlow in action
Switch
Data Path (Hardware)
OpenFlow
Any Host
OpenFlow Controller
OpenFlow Protocol
(SSL/TCP)
Host1 sends a packet
If there are no rules about handling this
packet
◦ Forward packet to the controller
◦ Controller installs a flow
Subsequent packets do not go through
the controller.
host1 host2
13

14. OpenFlow Basics
Flow Table Entries
Switch
Port
MAC
src
MAC
dst
Eth
type
VLAN
ID
IP
Src
IP
Dst
IP
ToS
TCP
sport
TCP
dport
Rule Action Stats
1. Forward packet to port(s)
2. Encapsulate and forward to controller
3. Drop packet
4. Send to normal processing pipeline
5. Modify Fields
Packet + byte counters
IP
Prot
VLAN
PCP
http://www.slideshare.net/Cameroon45/ppt-4515906
14

15. 1. SDN Background
2. Problem Statement
3. Motivation
4. Experimental Approach
5. Experimental Setup
6. Results
15

16. Problem Statement
 To build a stochastic model for end-to-end delay in OpenFlow enabled networks
based on the measurements and simulations on three platforms i.e. Mininet,
OFTEIN and GENI testbed.
16

17. 1. Background
2. Problem Statement
3. Motivation
4. Experimental Approach
5. Experimental Setup
6. Results
17

18. Motivation
The Interest in the accurate end-to-end delay measurement is twofold.
1. Deployment of real-time services necessitates delay constraints to be met.
2. From these end-to-end delay measurements we can learn about the underlying properties of the
network.
 So far very few studies have been conducted on the end-to-end delay analysis for OpenFlow enabled
networks.
 Only work done on Lab setup and network emulator Mininet.
 Need analysis to be done on the real networks. (GENI and OFTEIN testbed)
18

19. 1. Background
2. Problem Statement
3. Motivation
4. Experimental Approach
5. Experimental Setup
6. Results
19

20. Experimental Approach
Analyzed the performance of the OpenFlow enabled network.
Investigated the underlying parameters affecting the end to end delay in OpenFlow enabled
networks.
Observed the Internet traffic characteristics in OpenFlow enabled networks.
Developed reasonable model to understand these characteristics.
 Compare the performance for OpenFlow enabled networks on different platforms (i.e Mininet, GENI and
OFTEIN)
20

21. Literature Review
Jarschel, Michael, et al. "Modeling and performance evaluation of an openflow
architecture." Proceedings of the 23rd international teletraffic congress. International
Teletraffic Congress, 2011.
◦ Proposed a basic model for forwarding speed and blocking probabilities in an OpenFlow architecture
using queueing theory. Single switch model
◦ OMNeT++
Azodolmolky, Siamak, et al. "An analytical model for software defined networking: A network
calculus-based approach." Global Communications Conference (GLOBECOM), 2013 IEEE
◦ Delay and queue length boundaries are modeled using Network Calculus. Model only provided worst-
case bounds on performance metrics.
21

22. Literature Review
Chilwan, Ameen, et al. "ON MODELING CONTROLLER-SWITCH INTERACTION IN OPENFLOW
BASED SDNS.“ International Journal of Computer Networks & Communications (2014)
◦ A more accurate model using queueing theory but evaluated using simulations.
Bovy, C. J., et al. "Analysis of end-to-end delay measurements in Internet.“ Proceedings of
ACM Conference on Passive and Active Measurements (PAM), Fort Collins, Colorado, USA.
2007.
◦ A classification of the numerous histograms demonstrate that about 84% are typical histograms
possess Gamma-like shape with heavy tail.
22

23. 1. Background
2. Problem Statement
3. Motivation
4. Objectives
5. Experimental Setup
6. Results
23

24. Experimental Setups
SDN simulator- Mininet.
GENI testbed
OF@TEIN SDN Testbed.
24

25. Mininet
Mininet creates a realistic virtual network, running real kernel, switch and application code, on a single
machine (VM, cloud or native), in seconds, with a single command:
25

26. GENI Testbed
GENI1 (Global Environment for Network Innovations)
provides a virtual laboratory for networking and distributed
systems research and education.
26

27. GENI Testbed
GENI allows experimenters to:
•Obtain compute resources from locations around the United States.
•Connect compute resources using Layer 2 networks in topologies best suited to their experiments.
•Install custom software or even custom operating systems on these compute resources.
•Control how network switches in their experiment handle traffic flows.
27

28. OF@TEIN Testbed
28
OF@TEIN is a an OpenFlow enabled testbed spread over
nine countries.
Project was launched in July 2012, through Korean
Government funding.
Deployed on TEIN4 (Trans-Eurasia Information Network 4)
Managed by
◦ Consortium of Korean universities.
◦ International collaboration sites.
◦ Led by Gwangju Institute of Science & Technology (GIST), S.
Korea.

29. 1. Background
2. Problem Statement
3. Motivation
4. Objectives
5. Experimental Setup
6. Results
29

30. OF@TEIN
30

31. Experimental Setup
31
 Network topology consists of four switches.
 Two switches were at MYREN Site (Site in Malaysia) and other two switches were in PH Site
(Site in Philippine) and both the sites are connected through GRE tunnel.
 Controller was running at GIST Site (Site in Korea).
 We measured the Round-Trip Time (RTT), to avoid clock synchronization issues present in
measuring one-way delay.
 POX controller was used.
 OVS was used to enable OpenFlow 1.0.

32. 32
 Three scenarios were considered in experiments:
 Proactive - Controller populates the switch’s flow table ahead of time.
 Reactive - Switch does not find a flow table entry for an incoming round trip flow and
consults the controller.
 All to Controllers – All packets are forwarded through controller.
 Timeout value for a rule was set to 2 second.
 Total Packets= 10,000, Rate= 10 packets/s
 Three packet sizes were considered 40 bytes (small), 576 bytes (medium) and 1500 bytes
(large) for experiments.

33. VM
SMARTX BOX
GIST, Korea
VM
SMARTX BOX
PH Site, Philippine
SMARTX BOX
MYREN Site, Malaysia
GRE Tunnel

34. 34
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

• Proactive and All to Controller has Gaussian distribution on log scale.
• Reactive has a multimodal distribution.
• Lower end represents packets forwarded proactively
• Higher end represents packets forwarded through controller intervention
• Now lets see what happens if we change the packet size.

35. 35
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

• Packet size doesn’t have any effect the PDF of the either of the case.

36. 36
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

• Packet size doesn’t have any effect the PDF of the either of the case.

37. GENI – Case 1 – All
Controllers on same site
37

38. Experimental Setup
38
 Network topology consists of four switches.
 All switches were in KENTUCKY PKS2 Site (KENTUCKY State).
 4 Controllers were running at CENIC InstaGENI Site (California State).
 POX controller was used.
 OVS was used to enable OpenFlow 1.0.

39. Sender Receiver
KENTUCKY PKS2 Site
CENIC InstaGENI Site
29Mbps
29Mbps
29Mbps
29Mbps

40. 40
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers

Multiple site controllers
• Proactive and All to Controller has Gaussian distribution on log scale.
• Reactive has a multimodal distribution.
• Lower end represents packets forwarded proactively
• Higher end represents packets forwarded through controller intervention
• Now lets see what happens if we change the packet size.

41. 41
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers

Multiple site controllers
• Changing the Packet size changes the PDF of Reactive controller component.
• All other PDFs remains same.

42. 42
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers

Multiple site controllers
• Changing the Packet size changes the PDF of Reactive controller component.
• All other PDFs remains same.

43. GENI – Case 2 – All
Controllers on different sites
43

44. Experimental Setup
44
 Network topology consists of four switches.
 All switches were in KENTUCKY PKS2 Site (KENTUCKY State).
 4 different sites are selected for controllers i.e UCLA Site, Illinois Site, Ohio Site, and CENIC
Site.
 POX controller was used.
 OVS was used to enable OpenFlow 1.0.

45. Sender Receiver
KENTUCKY PKS2 Site
UCLA Site Illinois Site Ohio Site CENIC Site
28Mbps
59Mbps
63Mbps
29Mbps

46. 46
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers
Multiple sites controllers

• Proactive and All to Controller has Gaussian distribution on log scale.
• Reactive has a multimodal distribution.
• Lower end represents packets forwarded proactively
• Higher end represents packets forwarded through controller intervention
• Now lets see what happens if we change the packet size.

47. 47
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers
Multiple sites controllers

• Changing the Packet size have negligible effect the PDF of Reactive controller component.
• All other PDFs remains same.

48. 48
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

Single site controllers
Multiple sites controllers

• Changing the Packet size have negligible effect the PDF of Reactive controller component.
• All other PDFs remains same.

49. Mininet
49

50. Experimental Setup
50
 A linear topology with four switches created in mininet.
 POX controller was running on the same PC.
 OVS was used to enable OpenFlow 1.0.

51. 51
OVS Switch
Sender Receiver
Running Mininet on machine with Ubuntu 14.04
Mininet

52. 52
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

• Proactive has Gaussian distribution on log scale.
• All to Controller is a multimodal distribution.
• Reactive has a multimodal distribution.
• Lower end represents packets forwarded proactively
• Higher end represents packets forwarded through controller intervention
• Now lets see what happens if we change the packet size.

53. 53
• Packet size doesn’t have any effect the PDF of the either of the case.
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller


54. 54
OFTEIN GENI Mininet 40 576 1500
Proactive

Reactive
Switch part

Controller
part 
All to Controller

• Packet size doesn’t have any effect the PDF of the either of the case.

55. CDF based Estimation –
GENI – Case 1
55

56. Goodness-of-fit Criteria
56
Many criteria for information-based model selection have been devised in computational
learning theory, two best known are:
AIC, Akaike Information Criterion.
BIC, Bayesian Information Criterion.
The goal is to predict, using training data, which model has the best potential for accurate
generalization.

57. Akaike’s contribution (1973)
Akaike (1973) proposed “an information criterion” (AIC) (but now often called an Akaike
Information Criterion) that relates likelihood to K-L distance, and includes an explicit term for
model complexity…
K))y|(Lln(AIC 22  
This is an estimate of the expected,
relative distance between the fitted model
and the unknown true mechanism that
generated the observed data.
K=number of estimated parameters

58. General guidelines for use of AIC
We select the model with smallest value of AIC (i.e. closest to “truth”).
AIC will identify the best model in the set, even if all the models are poor!
It is the researcher’s (your) responsibility that the set of candidate models includes well
founded, realistic models.

59. Bayesian information criterion (BIC)
59
 The Bayesian information criterion (BIC) or Schwarz Criterion (also SBC, SBIC) is a criterion
for model selection among a class of parametric models with different numbers of
parameters.
 BIC is easy to calculate and enables us to approximate the marginal likelihood
n = number of data points
k = number of free parameters
RSS is the residual sum of squares
BIC = n ln(RSS/n) + k ln(n)
 We select the model with smallest value of BIC.

60. CDF estimation
GENI – Case 1
60

61. 61
Weibull Gamma Lognormal Normal
AIC 24525.48 15341.07 13436.53 20784.43
BIC 24539.90 15355.49 13450.95 20798.85
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers

62. 62
Weibull Gamma Lognormal Normal
AIC 23829.50 14881.91 13213.93 19638.29
BIC 23843.92 14896.33 13228.35 19652.71
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers

63. 63
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers
Weibull Gamma Lognormal Normal
AIC 26297.86 17535.03 15367.85 23385.74
BIC 26312.28 17549.45 15382.27 23400.16 Proactive case follows a lognormal distribution for 40 ,
576 and 1500 Byte.

64. 64
Weibull Gamma Lognormal Normal
AIC 19190.63 19150.43 19168.98 19123.82
BIC 19201.83 19161.63 19180.18 19135.02
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers

65. 65
Weibull Gamma Lognormal Normal
AIC 19181.34 19142.76 19159.46 19114.35
BIC 19221.65 19165.63 19191.45 19126.65
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers

66. 66
Weibull Gamma Lognormal Normal
AIC 19451.57 19111.36 19102.31 19090.97
BIC 19462.77 19122.56 19113.51 19070.17
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers

Multiple sites controllers
All to controller case follows a Normal distribution for
40 , 576 and 1500 Byte.

67. 67
Weibull Gamma Lognormal Normal
AIC 19028.43 12357.49 10874.11 16597.41
BIC 19042.28 12371.34 10887.96 16611.26
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers

Multiple sites controllers

68. 68
Weibull Gamma Lognormal Normal
AIC 20213.50 12893.10 11249.25 17564.29
BIC 20227.42 12907.02 11263.18 17578.21
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers

Multiple sites controllers

69. 69
Weibull Gamma Lognormal Normal
AIC 18415.34 11944.55 10650.63 15639.81
BIC 18429.17 11958.38 10664.46 15653.64
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers

Multiple sites controllers
Reactive (switch part) follows a Lognormal distribution
for 40 , 576 and 1500 Byte.

70. 70
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers

Multiple sites controllers
Weibull Gamma Lognormal Normal
AIC 26994.64 26712.90 26684.41 27070.26
BIC 27006.27 26724.53 26696.04 27081.89

71. 71
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers

Multiple sites controllers
Weibull Gamma Lognormal Normal
AIC 25429.37 25450.92 25395.83 25553.48
BIC 25440.77 25462.32 25407.23 25564.87

72. 72
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers

Multiple sites controllers
Weibull Gamma Lognormal Normal
AIC 27866.57 27482.20 27347.89 28068.74
BIC 27878.25 27493.88 27359.57 28080.42 Reactive (Controller part) follows a Lognormal
distribution for 40 , 576 and 1500 Byte.

73. CDF estimation
GENI – Case 2
73

74. 74
Weibull Gamma Lognormal Normal
AIC 24525.48 15341.07 13436.53 20784.43
BIC 24539.90 15355.49 13450.95 20798.85
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers


75. 75
Weibull Gamma Lognormal Normal
AIC 23829.50 14881.91 13213.93 19638.29
BIC 23843.92 14896.33 13228.35 19652.71
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers


76. 76
Weibull Gamma Lognormal Normal
AIC 26297.86 17535.03 15367.85 23385.74
BIC 26312.28 17549.45 15382.27 23400.16
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers

Proactive case follows a lognormal distribution for 40 ,
576 and 1500 Byte

77. 77
Weibull Gamma Lognormal Normal
AIC 19119.62 19027.04 19046.91 19003.08
BIC 19130.82 19038.24 19058.11 19014.28
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers


78. 78
Weibull Gamma Lognormal Normal
AIC 19389.42 19272.5 19287.9 19258.38
BIC 19400.62 19283.7 19299.1 19269.59
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers


79. 79
Weibull Gamma Lognormal Normal
AIC 19130.63 18998.83 19015.05 18981.94
BIC 19141.83 19010.03 19026.25 18993.15
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
Single site controllers
Multiple sites controllers

All to controller case follows a Normal distribution for
40 , 576 and 1500 Byte.

80. 80
Weibull Gamma Lognormal Normal
AIC 20890.98 13653.44 12058.81 18252.30
BIC 20905.00 13667.47 12072.84 18266.32
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers
Multiple sites controllers


81. 81
Weibull Gamma Lognormal Normal
AIC 21106.84 12943.96 11332.62 17609.03
BIC 21120.89 12958.00 11346.67 17623.08
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers
Multiple sites controllers


82. 82
Weibull Gamma Lognormal Normal
AIC 21007.68 13130.01 11613.59 17558.95
BIC 21021.72 13144.05 11627.63 17572.99
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Single site controllers
Multiple sites controllers

Reactive (switch part) follows a lognormal distribution
for 40 , 576 and 1500 Byte.

83. 83
Weibull Gamma Lognormal Normal
AIC 19445.12 19500.56 19675.48 19384.88
BIC 19456.07 19511.52 19686.44 19395.84
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers
Multiple sites controllers


84. 84
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers
Multiple sites controllers

Weibull Gamma Lognormal Normal
AIC 18430.65 18485.54 18676.46 18314.83
BIC 18410.34 18496.39 18687.31 18325.68

85. 85
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Single site controllers
Multiple sites controllers

Weibull Gamma Lognormal Normal
AIC 18786.28 18813.08 19020.67 18615.44
BIC 18797.17 18823.96 19031.55 18626.33 Reactive (controller part) follows a Normal distribution
for 40 , 576 and 1500 Byte.

86. CDF estimation
OFTEIN
86

87. 87
Weibull Gamma Lognormal Normal
AIC 52598.93 40778.92 40216.81 41924.08
BIC 52612.70 40792.70 40230.59 41937.86
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part

88. 88
Weibull Gamma Lognormal Normal
AIC 36246.37 31064.05 30718.17 31768.93
BIC 36259.26 31076.95 30731.06 31781.82
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part

89. 89
Weibull Gamma Lognormal Normal
AIC 46363.99 38522.75 38021.60 39440.25
BIC 46377.49 38536.24 38035.09 39453.74
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Proactive follows a lognormal distribution for 40 ,
576 and 1500 Byte.

90. 90
Weibull Gamma Lognormal Normal
AIC 18914.32 18828.94 18842.16 18801.01
BIC 18925.46 18840.09 18853.30 18810.15
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part

91. 91
Weibull Gamma Lognormal Normal
AIC 19662.66 18925.57 18945.92 18886.99
BIC 19673.83 18936.74 18957.09 18898.16
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part

92. 92
Weibull Gamma Lognormal Normal
AIC 15464.08 14790.32 14782.24 14507.56
BIC 15474.86 14801.10 14793.01 14518.33
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
All to controller case follows a Normal
distribution for 40 , 576 and 1500 Byte.

93. 93
Weibull Gamma Lognormal Normal
AIC 41617.03 32741.79 32293.84 33650.09
BIC 41630.32 32755.09 32307.14 33663.39
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part

94. 94
Weibull Gamma Lognormal Normal
AIC 38130.23 33544.40 33163.25 34320.94
BIC 38143.14 33557.31 33176.16 34333.85
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part

95. 95
Weibull Gamma Lognormal Normal
AIC 43439.00 35106.31 34791.09 35744.27
BIC 43452.48 35119.79 34804.57 35757.75
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Reactive (switch part) follows a lognormal
distribution for 40 , 576 and 1500 Byte.

96. 96
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 26470.65 26124.00 26041.68 26456.13
BIC 26481.92 26135.26 26052.94 26467.39

97. 97
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 22922.38 22663.10 22615.78 22938.80
BIC 22933.35 22674.06 22626.74 22949.76

98. 98
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 50292.76 49683.99 49514.51 50630.28
BIC 50305.18 49696.41 49526.94 50642.70
Reactive (controller component) follows a
lognormal distribution for 40 , 576 and 1500
Byte.

99. CDF estimation
Mininet
99

100. 100
Weibull Gamma Lognormal Normal
AIC 20721.28 1966.508 1603.499 2838.487
BIC 20735.71 1980.929 1617.920 2852.908
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part

101. 101
Weibull Gamma Lognormal Normal
AIC 21726.28 1980.123 1713.678 2878.543
BIC 20875.71 1995.999 1705.135 2879.873
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part

102. 102
Weibull Gamma Lognormal Normal
AIC 20721.28 1966.508 1603.499 2838.487
BIC 20735.71 1980.929 1617.920 2852.908
GENI OFTEIN Mininet 40 576 1500
Proactive

All to Controller
Reactive
Switch part
Controller
part
Proactive follows a lognormal distribution for 40
, 576 and 1500 Byte.

103. 103
Weibull Gamma Lognormal Normal
AIC 16338.03 16425.39 16521.57 16421.56
BIC 16349.23 16436.59 16532.77 16432.76
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part

104. 104
Weibull Gamma Lognormal Normal
AIC 16315.45 16476.32 16621.53 16435.35
BIC 16367.32 16464.43 16665.34 16425.65
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part

105. 105
Weibull Gamma Lognormal Normal
AIC 16331.96 16420.89 16516.72 16413.81
BIC 16343.16 16432.09 16527.92 16425.01
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller

Reactive
Switch part
Controller
part
All to controller case follows a Weibull
distribution for 40 , 576 and 1500 Byte.

106. 106
Weibull Gamma Lognormal Normal
AIC 370206.2 2802.091 1543.293 5525.703
BIC 370220.6 2816.426 1557.628 5540.038
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part

107. 107
Weibull Gamma Lognormal Normal
AIC 415560.4 1705.901 1341.1055 4630.861
BIC 415574.8 1720.236 1355.4406 4645.196
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part

108. 108
Weibull Gamma Lognormal Normal
AIC 20721.28 1966.508 1603.499 2838.487
BIC 20735.71 1980.929 1617.920 2852.908
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part

Controller
part
Reactive (switch part) case follows a Lognormal
distribution for 40 , 576 and 1500 Byte

109. 109
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 3426.519 3451.025 3471.003 3442.220
BIC 3434.594 3459.101 3479.079 3450.295

110. 110
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 3412.248 3444.611 3467.717 3428.952
BIC 3420.314 3452.677 3475.783 3437.018

111. 111
GENI OFTEIN Mininet 40 576 1500
Proactive
All to Controller
Reactive
Switch part
Controller
part 
Weibull Gamma Lognormal Normal
AIC 3444.185 3464.891 3483.08 3459.23
BIC 3452.275 3472.981 3491.17 3467.32
Reactive (controller part) case follows a Weibull
distribution for 40 , 576 and 1500 Byte

112. Stochastic Analysis of Results
112
We model the end-to-end delay in an OpenFlow SDN network, as a sum of two components:
Deterministic delay ( ) and stochastic delay ( ), i.e.𝐷 𝑑 𝐷𝑠
𝐷 𝐸2𝐸 = 𝐷 𝑑 + 𝐷𝑠
These two terms are further decomposed in terms of the
following equation:
(1)
𝑫 𝑬𝟐𝑬 =
𝒊=𝟏
𝒍
𝑫 𝒕𝒓𝒂𝒏𝒔,𝒊 + 𝑫 𝒑𝒓𝒐𝒑,𝒊 +
𝒊=𝟏
𝒏 𝑺
𝑺 𝒔,𝒊 +
𝒋=𝟏
𝒏 𝑪
𝑰𝒋 × 𝑺 𝒄,𝒋 (2)

113. Stochastic Analysis of Results
113
In Equation 2, 𝑫 𝒑𝒓𝒐𝒑,𝒊 is the propagation delays of the link on the path between sender and
receiver, each of which is calculated as
𝑫 𝒕𝒓𝒂𝒏𝒔,𝒊 is the transmission delay over the link between sender and receiver
The terms 𝑺 𝒔,𝒊 and 𝑺 𝒄,𝒋 in Equation 2 are the stochastic delays associated with the 𝑖 𝑡ℎ switch
and controller, respectively.
𝑫 𝒑𝒓𝒐𝒑,𝒊 =
Distance𝒊
Speed𝒊
𝑫 𝒕𝒓𝒂𝒏𝒔,𝒊 =
Number of bits
Link transmission rate
𝑖 𝑡ℎ
𝑖 𝑡ℎ
𝑗𝑡ℎ
𝒏 𝑺 = total no. of switch on the path
𝒏 𝑪 = max. no. of controllers in the path

114. Stochastic Analysis of Results
114
are Bernoulli random variables that take on value 1 with probability and value 0 with
probability , also called an Indicator function.
The values of 𝛼𝑗 depend on a variety of factors including the timeout value of flow table
entries in switches, input traffic rate, no. of the controllers and placement of the controllers.
𝑫 𝑬𝟐𝑬 =
𝒊=𝟏
𝒍
𝑫 𝒕𝒓𝒂𝒏𝒔,𝒊 + 𝑫 𝒑𝒓𝒐𝒑,𝒊 +
𝒊=𝟏
𝒏 𝑺
𝑺 𝒔,𝒊 +
𝒋=𝟏
𝒏 𝑪
𝑰𝒋 × 𝑺 𝒄,𝒋
𝐼𝑗 𝛼𝑗
1 − 𝛼𝑗

115. Observations
115
 It has been found the PDF of end-to-end delay in OpenFlow switch SDNs is multi-modal
rather than unimodal distributions which is the case for traditional networks.
 It is observed that PDF for the proactive case is Log-normal for all the platforms (i.e.
OFTEIN, GENI and Mininet)
 For All to controller case PDF is Normal for OFTEIN and GENI and Weibull for Mininet.
 For Reactive GENI case 1 and OFTEIN (co-located controllers ) PDF is lognormal for
controller component .
 For Reactive GENI case 2 (distributed controllers) PDF is Normal for controller component.
 For Reactive Mininet PDF is Weibull for controller component.
 For Reactive GENI, OFTEIN and Mininet case PDF is Lognormal for switch component.

116. Thank You
116