CS4554 Network Modeling

1
CS4554 Network Modeling Design
ProjectAvailability Analysis and Simulation

• Lt Com. Len Gaines
• Capt. Dag-Anders Brunstad
• Ltjg. Ugur Demiryurek
• 1Lt. Mohammad Ababneh

2
Analytical Model

• We have decided to do the analysis of only a part
  of the Norwegian Network for the purpose of our
  project.
• We were dealing with 6 nodes and 7 links
• We will calculate the availability and
  reliability of the network

3
Definitions

• Availability measures a systems ability to be in
  place and working when it is needed.
• In a network, system availability is determined
  by the reliability of the components that make up
  the system, the maintenance philosophy, the
  ability of logistically support the system, and
  the redundancy built into the system
• Availability takes into account the time needed
  to repair a system when it has failed.

4
Definitions

• Reliability measures how long a system will
  continue to perform.
• Reliability analysis is only concerned with the
  probability that a failure will occur when the
  system is needed.
• It does not take into account downtime or the
  time to repair.

5
Definitions

• Failure Rate (?) is the random arrival rate of
  errors.
• ? ?t, where ? is the number of arrivals over a
  period of time t.

6
Definitions

• MTTR Mean time to repair
• MTBE Mean time between events
• MLDT Mean logistic delay time.
• Recovery Rate (?) It is the rate taken to
  recover from an error.
• ? 1/MTTR. It is also represented by the total
  number of events/total time the system was down.

7
The Goal

• Calculate the overall network availability
• Answer the question How can we improve the
  availability?
• Strengthen the nodes
• Add a new link

8
Parameters

• ? the number of arrivals over a period of time
  t
• t a period of time.
• MTTR Mean Time To Repair

9
Assumptions

• The number 6 router had failed 25 times in 1,000
  hours
• The failure rates are similar for all routers
• Link F had 10 failures over the last 1,000 hours
• Each link had a similar failure rate
• The events are independent and identically
  distributed.
• Failures will occur according to a Poisson
  distribution
• A recovery rate (?) of 0.4 for the routers and a
  ? of 0.2 for the links

10
Metrics

• Availability of the network
• Reliability of the network
• Probability of zero errors on each router and link

11
The Analysis

• The analysis is done in two steps
• Step 1 Calculate the probability of zero errors
  for the routers and links
• Step 2 Calculate the overall network availability

12
The probability of zero errors for the routers
and links

• In a Poisson distribution, the probability of a
  failure is Pk(t) e-?t (?t)k/k!.
• Pk(t) represents the probability of k failures
  during any particular time interval of length t
• Using our assumptions ? and t
• probability of zero errors
• For routers 0.97
• For links 0.99

13
The Network Analysis

• The reliability of the network is calculated by
  finding the aggregate probability of zero errors
  for all the routers and links

14
Probability Background

• Based on the assumption of independence.
• The probability that a system will work if it
  consists of nodes in series is
• P(A ? B) P(A) P(B)
• ExP(A) 0.9, P(B) 0.9, P(A ? B) 0.81

15
Probability Background

• The probability that a system will work if it
  consists of nodes in parallel is
• P(A ? B) P(A) P(B) (P(A) P(B))
• ExP(A) 0.9, P(B) 0.9,
• P(A ? B) 0.9 0.9 0.81 0.99

A
B
16
Probability Background

• Ex For the system shown
• The probability that a system will work if P(A)
  P(B) P(C) P(D) 0.9 is
• P(A ? B) ? P(C ? D) 0.81 0.81-(0.810.81)
  0.9639

17
The Network Analysis

• For our network, we represented each router and
  each link as a node in a system
• We calculated the probability that the system
  will work when needed with zero errors

18
The Network Analysis
A
1
2

• The Network

C
D
B
3
4
E
F
5
G
6
19
The Network Analysis

• The probability that the system will work is
• P(1) ? (P(A) ? P(2) ? P(D) ? (P(C) ? P(3) ?
  P(E))) ? (P(B) ? P(3) ? P(E) ? (P(C) ? P(2) ?
  P(D))) ? P(4) ? P(F) ? P(6) ? P(G) ? P(5)

20
The Network Analysis
P(1) (P(A) P(2) P(D) (P(C) P(3)
P(E)) (P(D) P(C) P(3) P(E))) (P(B)
P(3) P(E) (P(C) P(2) P(D)) (P(E)
P(C) P(2) P(D))) (P(A) P(2) P(D)
(P(C) P(3) P(E)) (P(D) P(C) P(3)
P(E))) (P(B) P(3) P(E) (P(C) P(2)
P(D)) (P(E) P(C) P(2) P(D))) P(4)
P(F) P(6) P(G) P(5)
21
The Network Analysis
.97 (.99 .97 .99 (.992 .97) (.993
.97)) (.99 .97 .99 (.992 .97) (.993
.97)) - (.99 .97 .99 (.992 .97)
(.993 .97)) (.99 .97 .99 (.992 .97)
(.993 .97)) .973 .992 .97 (.9603
.99 (.9506) (.9412)) (.9603 .99
(.9506) (.9412)) - (.9603 .99 (.9506)
(.9412)) (.9603 .99 (.9506) (.9412))
.91 .98 .97 (.9603 .9995) (.9603
.9995) - (.9603 .9995) (.9603
.9995) .89 .97(.96 .96 - .962) .89
0.86
22
The Network Analysis

• The probability that the system will function at
  any given time is 0.86.
• This figure does not account for the time that a
  system is down due to preventive maintenance or
  failure.
• The figure 0.86 represents reliability, not
  availability.
• If the analytical model was correct, the
  availability results from simulations should be
  below 0.86

23
The Network Availability Simulation
24
The Simulation Tool MeaDep

• MeaDep MEAsure DEPendability
• MEADEP consists of four modules
• the Data Pre-Processor (DPP)
• the Data Editor and Analyzer Module (DEA).
• the Model Generator Module (MG).
• the Model Evaluator (ME).
• Specialized availability and reliability tool.

25
Data Evaluator and Analyzer
26
Query Form
27
Event Pie Chart
28
Duration Pie Chart
29
Router Statistics
30
Router Statistics
31
Router Statistics
32
Router Statistics
33
Network Model in Model Generator Module
34
Network ModelDrilling down
35
Network Model
36
Network Model
37
Network Model
38
MEADEP Results
Model-Name Failure-Rate Recovery-Rate
Availability Unavailability
(per hour) (per hour)
-------------------------------------------------
------------------------------- Norway
0.1442698363 0.3416837669 0.7031201428
0.2968798572 Expand 0.00376564428
0.3080840826 0.9879248114 0.01207518865 Expand
A 0.03803392613 0.3080840826
0.8901128367 0.1098871633 ExpandB
0.03803392613 0.3080840826 0.8901128367
0.1098871633 Expand2 0.00182121084
0.2595515987 0.9930321335 0.006967866512 Expan
dC3 0.04746634615 0.2769230769
0.8536748033 0.1463251967 Expand3
0.00182121084 0.2595515987 0.9930321335
0.006967866512 ExpandC2 0.04746634615
0.2769230769 0.8536748033 0.1463251967
39
6 Sensitivity Analysis for µ
40
2 Sensitivity Analysis for µ
41
Change µ to 1
42
New Network Results
43
Non-Repair Scenario
44
Validated Analytical Model

• The Failure Rate or ? is .15
• Probability of zero errors in a Poisson
  distribution Pk(t) e-?t (?t)k/k!
• Remember that ? ?t
• So e-.15 .150/0! e-.15 .86
• The same reliability rate as the analytical model
• We could not validate their reliability figure,
  because we did not know the distribution and it
  was over 100 hours instead of 1.

45
MEADEP

• Powerful, easy to use tool
• Good sensitivity analysis
• ME module allows easy editing to perform numerous
  permutations
• Modeling can be difficult and hard to analyze
• Prior knowledge of statistics and probabilities
  is required

46
Extend

• Powerful, easy to start using
• You have full flexibility
• Very effective when modeling simple functions
• Nice animation features

47
(No Transcript)
48
Inside Router Module
49
Inside Channel Module
50
Using Excel to Monitor
51
Extend Conclusion

• The Extend model was not completed fare eunuch to
  get comparable results
• Time consuming modeling technique because you
  must make every simulation function from basic
  building blocks
• Build up your own libraries with modules as you
  are gaining experience

52
OPNET Model Overview

• GOAL
• Reducing network delay by adding a new link
• CRITERIA
• Overall Delay
• Load
• Router CPU Utilization
• Router Processing Delay

53
Simulation Info

• Simulated 1 hour
• Events 651364
• Average speed 10974 event/sec
• Time 58 sec on 400 Mhz Pentium

54
Norwegian Network
55
Delay
56
Load on Server1 and Server2
57
Add new T1 Line
58
Delay with new T1
59
Load with new T1 line
60
CPU Utilization with new T1
61
Router Processing Delay w T1
62
Conclusion

• Overall Delay decreases
• As expected Load on Server2 increases but not
  significant
• Router CPU utilization decreases
• Router Processing Delay is almost same
• It is feasible to add new T1 line between R1
  and R6

63
About OPNET

• CONS
• No reliability or availability analysis is
  encountered
• Not easy to use compare to similar tools
• Many bugs but not mentioned by the company
• PROS
• Lets user to import initial topology from HP
  Openview, etc