1. A PATTERN MATCHING APPROACH
TO CLASSIFICATION OF INTERNET TRAFFIC
Submitted By
P.SAKTHIVEL(Reg.No. 2010PR197)
Under the Guidance of
C.SENTHILKUMAR M.Sc.,M.Phil.,
Associate Professor, Dept. of Computer Science,
Erode Arts & Science College,Erode-638009.
2. Objective of Study
Different types Online and Offline Tools were developed to
understanding the character of internet traffic types.
New approaches are to identify proposed traffic modeling.
Data sets were collected from different organizations and
different types of networks.
3. Abstract
The past network traffic used self-similarity particularly
linear predication based traffic model.
Auto regressive / ARMA models have been used monitor
the Network traffic.
The traffic type can be characterized by Packet length,
Packet inter arrival time, Connection direction, Connection
Packet count and byte count.
4. The Proposed model used mixture distribution under
simple K-means clustering based approach
To facilitate the traffic analysis, developed several tools.
5. Literature Review on traffic modeling
Existing Model:
Linear Predication based traffic model
LP based network traffic uses the past history of network
traffic to predict the future.
This model have a minimal number of parameters and
overhead for modeling.
It identify the optimal value of LP parameters for network
traffic
6. The LP model can predict the occurrence of network
congestion.
LP model by itself does not lead to the source of
congestion.
7. Need for a New model
LP based traffic model can predict a network fault but
cannot find the source of the network fault.
A new model is required to overcome the limitation of
recent LP-based network traffic model.
8. Tools Developed and Data Sets Collected
Offline Tool
Online Tool
WANMon Tool
Control Tool
Data Sets
TeNeT-WAN-2005
TeNeT-LAN-2005
BHEL-WAN-2005
9. UNIVARIATE DISTRIBUTION MODELING
Fitting Univariate Distributions to Internet Tra_c
Types:
Analysis can be divided into the following steps:
Identifying the shape of the density
Estimating the unknown parameters
Performing a test of fit
10. Identifying the Shape of the Density
For every member of this system, the probability density
fx(x) satises the differential equation.
Estimating the Unknown Parameters
The general form of density function for the Type I family
of distributions is given below:
11. Performing a Test of Fit
Under the condition that there are k intervals of values of x
giving nonzero expected frequencies. The parameters in
equation (4.17) are as follows:
• nfe(x) the expected frequencies,
• foi the observed frequencies,
• n the total number of samples.
12. MIXTURE DISTRIBUTION MODELING
K-Means Clustering Based Approach
Training:
A series of K-means clustering models are trained to
model the traffic types from the available training data.
Testing:
Minimum Distance Computation (MDC) identifies the
traffic type by calculating the distance of an unknown
traffic type to these models
23. CONCLUSION
Easily access the internet to the user.
High bandwidth.
Variety of off and on line tools are used.
Easily identify the traffic model.