![](bg4.png)
PaxRunner (PaxRunner, 2009) ease the management
of bundle provision inside interchangeable OSGi
framework implementations. Managing bundles life
cycle (install, run, stop, uninstall) is also simple.
Another important benefit of the OSGi
technology is that Cisco network equipments already
integrate an implementation of the OSGi platform
(Cisco, 2008). Remote deployment of OSGi bundles
is the only thing to do to integrate the autonomic
network management system in a Cisco piece of
equipment.
3.1.2 Web Ontology Language and Semantic
Web Rule Language
A cognitive network manager needs to have a
“world” representation and capture the technical
know-how in order to work autonomously with
respect to the operator objectives. Therefore, a tool
that allows the network expert to express its
knowledge in a machine-readable way is required.
An ontology is a « shared and common
understanding of a domain that can be
communicated between people and heterogeneous
and distributed systems » (Fensel, 2001). It enables
representing domain background knowledge in a
machine understandable form (Studer, 1998). Using
such a formal model within our distributed
architecture is appropriate. An ontology defines a set
of concepts, properties, relationships, constraints and
axioms that provide rules that govern them.
The Web Ontology Language (OWL) (W3C,
2004) is the W3C standard for ontological
modelling. It has been designed to provide a
common way to process the content of web
information. The OWL standard defines three
increasingly expressive dialects: OWL Lite, OWL
DL and OWL Full. OWL Full contains all the OWL
language constructs but has no computation
guarantees because it introduces too many
possibilities. OWL DL is a sublanguage of OWL
Full and relies mostly on description logics (DL).
OWL DL is computationally decidable and more
appropriate for knowledge representation when
inference is needed. OWL Lite is a subset of OWL
DL and suits well for expressing basic classification
hierarchy and simple constraints. Although
originally defined as an important part of the
semantic Web suite, OWL is emerging as the major
standard for knowledge representation.
However, OWL constructs do not allow the
formalization of rules on top of the ontology.
Among many proposals aiming at enhancing OWL
knowledge bases with rules, the Semantic Web Rule
Language (SWRL) (W3C, 2004) is probably the best
known and most established. SWRL provides the
means to define rules that extends the OWL set of
axioms.
3.1.3 Jess Inference Engine
Cognitive network managers need reasoning
capabilities to make decisions according to policies
defined by network administrators. An inference
engine performs reasoning from declarative facts.
Jess, for Java Expert System Shell (Friedman-Hill,
2003), is a fast and powerful rule engine for the Java
platform, which supports development of rule-based
systems that can be tightly coupled to code entirely
written in Java. Jess has been integrated with several
agent frameworks and other tools like the popular
ontology editor Protégé (Protégé, 2009). Jess, which
supports both forward and backward chaining, has
been integrated in the CONEMAF platform to
provide such reasoning capabilities.
3.2 Modular Framework
CONEMAF is built on top of OSGi and follows the
modular principles that OSGi enables. Software
upgrade, deployment over heterogeneous network
elements are thus facilitated.
3.2.1 Framework Components Overview
Figure 2 represents the different components the
CONEMAF platform is made of from a software
point of view. The core framework is composed of
components that are essential for the cognitive
network manager to play its role. This includes a
scheduler to trigger the execution of behaviours, an
inference engine for decision-making, and a
blackboard, which acts as an organized common
space for information sharing. Topology, discovery
and communication services are also implemented
as modules. The main benefit of the framework
resides in the simplicity of adding, deleting or
changing one of its components. Behaviours and
network element controllers that may be adapted to
the type of device they are embedded in particularly
aim at exploiting such a benefit. All these
components are individually described in the
following section.
3.2.2 Modules Description
Each component of the cognitive network manager
is implemented as a module, called bundle in the
OSGi terminology. The present release of
CONEMAF: A MODULAR MULTI AGENT FRAMEWORK FOR AUTONOMIC NETWORK MANAGEMENT
227