17-03-2014, 08:03 PM
Abstract—IP traffic or Internet flow classification has played
an important role in many network activities. Many works
have been carried out for the past several years targeting
different goals including Quality-of-Service support, security,
management, and provisioning. However, these approaches are
developed and evaluated separately without a general model that
can be used to identify, categorize, and compare them. Thus, it
is difficult for a newcomer to understand the principle of flow
classification and to find a suitable method for her domain. This
paper brings forth a mathematical model that describes the flow
classification components and processes on an abstract level. We
then classify existing flow classification approaches and compare
them in a unified framework.
I. INTRODUCTION
IP traffic or Internet flow classification is a method to
categorize a given flow to an appropriate group or class. An
implementation of such method is called a flow classification
system. It has played an important role in many areas and
several flow classification systems have been proposed. Some
are designed for network management and administration
[8][13][4][10][11], some for QoS support [12][19][18][1][16],
and some for assisting the network security [9][20][3]. However,
different systems are developed independently without
any general underlying model that can explain the common
components and processes, making analyzing, comparing, and
understanding the relationships among different approaches
rather difficult. This issue has also been raised in [15] but the
authors have only introduced a reliable measurement method,
not a model that can identify the classification systems. Therefore,
in this paper, we propose a mathematical model that can
describe the flow classification components and processes as
well as existing classification approaches. The model provides
a unified framework for analysis and comparison of different
flow classification approaches. The components and processes
are formulated using mathematical constructs which are clear,
formal, and well-defined. The existing flow configuration
systems are modeled and divided into four categories wrt.
their abilities to update their classifiers automatically and to
classify flows in real-time. The framework is formally modeled
in Section II. In Section III, several existing classification
approaches are modeled, categorized and compared under
a unified framework. This shows that, regardless of how
complicated the classification approaches are, they can be
formulated using our model.
an important role in many network activities. Many works
have been carried out for the past several years targeting
different goals including Quality-of-Service support, security,
management, and provisioning. However, these approaches are
developed and evaluated separately without a general model that
can be used to identify, categorize, and compare them. Thus, it
is difficult for a newcomer to understand the principle of flow
classification and to find a suitable method for her domain. This
paper brings forth a mathematical model that describes the flow
classification components and processes on an abstract level. We
then classify existing flow classification approaches and compare
them in a unified framework.
I. INTRODUCTION
IP traffic or Internet flow classification is a method to
categorize a given flow to an appropriate group or class. An
implementation of such method is called a flow classification
system. It has played an important role in many areas and
several flow classification systems have been proposed. Some
are designed for network management and administration
[8][13][4][10][11], some for QoS support [12][19][18][1][16],
and some for assisting the network security [9][20][3]. However,
different systems are developed independently without
any general underlying model that can explain the common
components and processes, making analyzing, comparing, and
understanding the relationships among different approaches
rather difficult. This issue has also been raised in [15] but the
authors have only introduced a reliable measurement method,
not a model that can identify the classification systems. Therefore,
in this paper, we propose a mathematical model that can
describe the flow classification components and processes as
well as existing classification approaches. The model provides
a unified framework for analysis and comparison of different
flow classification approaches. The components and processes
are formulated using mathematical constructs which are clear,
formal, and well-defined. The existing flow configuration
systems are modeled and divided into four categories wrt.
their abilities to update their classifiers automatically and to
classify flows in real-time. The framework is formally modeled
in Section II. In Section III, several existing classification
approaches are modeled, categorized and compared under
a unified framework. This shows that, regardless of how
complicated the classification approaches are, they can be
formulated using our model.