Comparing the Network Modeling Techniques Ivan Kendzor, Max Helm - - PowerPoint PPT Presentation

Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Comparing the Network Modeling Techniques

Ivan Kendzor, Max Helm

Friday 25th January, 2019 Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Introduction

Purpose of Network Modeling Techniques Advantages of using Network Modeling Techniques

• Measuring of network performance in the real systems
• Monitoring and obtaining precise metrics of the network behavior
• Predicting of impact of changes and architecture decision on flow performance
• Avoids disruption in the real system, enabling the assessment of new processes
• Modeling promotes and improves learning

Network modeling techniques can evaluate such performance metrics:

• Throughput
• Waiting time
• Estimation the delay of networks
• Packet processing time and packet loss

I.Kendzor — Comparing the Network Modeling Techniques 2

Introduction

Classification of Network Modeling Techniques

• Analytical models
• Hidden Markov Model
• Network Calculus

I.Kendzor — Comparing the Network Modeling Techniques 3

Introduction

Classification of Network Modeling Techniques

• Analytical models
• Hidden Markov Model
• Network Calculus
• Machine Learning Models

I.Kendzor — Comparing the Network Modeling Techniques 3

Introduction

Classification of Network Modeling Techniques

• Analytical models
• Hidden Markov Model
• Network Calculus
• Machine Learning Models
• Artificial Neural Network Models
• Multilayer perceptron neural network model
• Radial Basis Function Networks
• State-Space Dynamic Neural Network Technique

I.Kendzor — Comparing the Network Modeling Techniques 3

Analytical Models

Hidden Markov Model

• Defines the system of markov process consisting of hidden states.
• Core method: Markov Model
• Application field: Coverage problem in wireless sensor networks with two goals:
• Energy efficiency
• Connectivity

1 1P

. Chaturvedi und A. K. Daniel, „Hidden markov model based node status prediction technique for target coverage in wireless sensor networks“, in 2017 International Conference on Intelligent Communication and Computational Techniques (ICCT), Jaipur,2017, S. 223–227.

I.Kendzor — Comparing the Network Modeling Techniques 4

Analytical Models

Wireless Sensor Networks A network of devices that can communicate the information gathered from a monitored field through wireless links

Figure 1: Wireless Sensor Networks2

• 2Y. El Khamlichi, A. Tahiri, A. Abtoy, I. Medina-Bulo, and F

. Palomo-Lozano, “A Hybrid Algorithm for Optimal Wireless Sensor Network Deployment with the Minimum Number of Sensor Nodes,” Algorithms, vol. 10, no. 3, p. 80, Jul. 2017.

I.Kendzor — Comparing the Network Modeling Techniques 5

Analytical Models

Markov Model

• States {S1, S2, S3}
• State transitions probability pij = P[tj(t + 1) | ti(t)]0 ≤ i, j ≤ n
• Matrix representing the probability of states P =
• 0.4

0.6 0.7 0.3

• Initial distribution π =

0.3 0.7

• Markov Model enables to calculate the probability of sequence of states

Figure 2: Example of states

• Probability of sequence: Sunny - Rainy - Rainy - Sunny: 0.3 ∗ 0.6 ∗ 0.3 ∗ 0.7 = 0.0378

I.Kendzor — Comparing the Network Modeling Techniques 6

Analytical Models

Markov Model

Figure 3: Probability of observation sequence of lenght 3 3

3P

. Chaturvedi und A. K. Daniel, „Hidden markov model based node status prediction technique for target coverage in wireless sensor networks“, in 2017 International Conference on Intelligent Communication and Computational Techniques (ICCT), Jaipur,2017, S. 223–227.

I.Kendzor — Comparing the Network Modeling Techniques 7

Analytical Models

Hidden Markov Model

• Observation probability matrix bj(k) = P[ak (t) | tj(t)]0 ≤ j ≤ n, 0 ≤ k ≤ m
• Matrix representing the probability of observations P =
• 0.1

0.5 0.4 0.6 0.3 0.1

• Figure 4: Example of observations
• Observation sequence probability

P(x, a) = πa0qa0(a0)pa0a1qa1(a1)pa1a2qa2(a2)

Figure 5: Probability of states at various places for observation sequence of length 3

I.Kendzor — Comparing the Network Modeling Techniques 8

Analytical Models

Network Calculus

• A theory of deterministic queuing systems for the internet.
• Core method: Cumulative Rate Function
• Application field:
• Performance analysis of Network Coding4
• Reduction of data transmission amount in wireless sensor networks 5
• 4H. Li, X. Liu, W. He, J. Li, und W. Dou, „End-to-End Delay Analysis in Wireless Network Coding: A Network Calculus-Based Approach“, in 2011 31st International Conference on Distributed

Computing Systems, Minneapolis, MN, USA, 2011, S. 47–56.

• 5L. Jiang, L. Yu, und Z. Chen, „Network calculus based QoS analysis of network coding in Cluster-tree wireless sensor network“, in 2012 Computing, Communications and Applications

Conference, Hong Kong, China, 2012, S. 1–6.

I.Kendzor — Comparing the Network Modeling Techniques 9

Analytical Models

Network Coding Example

Figure 6: Example of Packet Switch Method Figure 7: Example of Network Coding

I.Kendzor — Comparing the Network Modeling Techniques 10

Analytical Models

Network Calculus Main components:

• Arrival curve
• Service curve

Figure 8: Example of Network Calculus6

• 6H. Li, X. Liu, W. He, J. Li, und W. Dou, „End-to-End Delay Analysis in Wireless Network Coding: A Network Calculus-Based Approach“, in 2011 31st International Conference on Distributed

Computing Systems, Minneapolis, MN, USA, 2011, S. 47–56.

I.Kendzor — Comparing the Network Modeling Techniques 11

Machine Learning Network Models

Machine Learning Modeling Techniques

• Core method: machine-learning algorithms
• Application field:network optimization problem; estimating quality of transmissions

Machine Learning Models are based on two enabling technologies:

• Software-defined networking
• Network analytics

Most applied machine learning algorithms:

• Linear regression
• Linear discriminant analysis
• Naive bayes
• k-nearest neighbors

I.Kendzor — Comparing the Network Modeling Techniques 12

Machine Learning Network Models

Machine Learning Modeling Techniques

Figure 9: Example of "what-if" Machine Learning Algorithms7

7F

. Geyer und G. Carle, „Towards automatic performance optimization of networks using machine learning“, in 2016 17th International Telecommunications Network Strategy and Planning Symposium (Networks), Montreal, QC, Canada, 2016, S. 19–24.

I.Kendzor — Comparing the Network Modeling Techniques 13

Artificial Neural Networks

Artificial Neural Networks Models

• Core method: backpropagation algorithms
• Application field:
• Electromagnetic simulation
• Modeling of power amplifiers and circuits
• Modeling input-output relationship

Structure:

• Main element of artificial neural network is neuron
• ANN and is composed of three components
• a set of connecting links (synapses)
• cumulative function for summing the input signals
• activation function to control the amplitude of the output
• Neurons consist of the following layers:
• one Input layer
• one Output layer
• one or many Hidden layers

I.Kendzor — Comparing the Network Modeling Techniques 14

Artificial Neural Networks

Method of backpropagation

Figure 10: Overview of Artificial Neural Network Model 8

I.Kendzor — Comparing the Network Modeling Techniques 15

Artificial Neural Networks

General Overview of Network Modeling Techniques

Figure 11: General Summary of Network Modeling Techniques

I.Kendzor — Comparing the Network Modeling Techniques 16

Bibliography

• F. Geyer und G. Carle, „Towards automatic performance optimization of networks using machine learning“, in 2016 17th Interna-

tional Telecommunications Network Strategy and Planning Symposium (Networks), Montreal, QC, Canada, 2016, S. 19–24. P . Chaturvedi und A. K. Daniel, „Hidden markov model based node status prediction technique for target coverage in wireless sensor networks“, in 2017 International Conference on Intelligent Communication and Computational Techniques (ICCT), Jaipur, 2017, S. 223–227.

• S. Gebert, T. Zinner, S. Lange, C. Schwartz, und P

. Tran-Gia, „Performance Modeling of Softwarized Network Functions Using Discrete-Time Analysis“, in 2016 28th International Teletraffic Congress (ITC 28), Würzburg, Germany, 2016, S. 234–242.

• H. Li, X. Liu, W. He, J. Li, und W. Dou, „End-to-End Delay Analysis in Wireless Network Coding: A Network Calculus-Based

Approach“, in 2011 31st International Conference on Distributed Computing Systems, Minneapolis, MN, USA, 2011, S. 47–56.

• D. E. Rumelhart, G. E. Hinton, und R. J. Williams, „Learning representations by back-propagating errors“, Nature, Bd. 323, S.

533, Okt. 1986.

• S. Yan, W. Shi, und J. Wen, „Review of neural network technique for modeling PA memory effect“, in 2016 IEEE MTT-S Interna-

tional Conference on Numerical Electromagnetic and Multiphysics Modeling and Optimization (NEMO), Beijing, China, 2016, S. 1–2.

I.Kendzor — Comparing the Network Modeling Techniques 17