Work package 3:
The piloting plane
Workpackage chair | UFRJ |
Partners | Devoteam, LIP6, Netcenter, Ginkgo, PUC, Telecom SudParis, Unicamp |
The work
package objective is to define the piloting
system necessary for handling all the problems arising in the network:
congestion, failure, QoS problem, etc. The first task will be oriented
on the
knowledge. The second task will focus on how the piloting plane is
recovering
the best possible knowledge adapted to the control and managements
algorithms.
The process for deciding what knowledge has to be recovered, where to
find this
knowledge and when to retrieve this knowledge will also be studied.
Networks represent a very dynamic and complex area, in
which every day managers are facing new problems and challenges.
Increases in
network complexity and information volume make resources and network
control more
and more difficult. We argue that in such an unpredictable, changing
and open
environment, intelligent agents give the opportunity to obtain an
optimized
network management and control. In fact, the main features of agents,
namely
the autonomy, the ability to communicate with the others, the ability
to solve common
problems in a decentralized manner, the ability to cooperate. In
addition some
learning aptitudes, should allow agents to operate in network dynamic
environment.
In this project, we are interested in adaptive
management and control approaches based on agents which monitor the
network
state and control its different components. Our aim is to include some
intelligent and dynamic control, thanks to agents, allowing us to
guarantee a
QoS and to give a better management and global performance of the
network.
The monitoring that could be
performed using
intelligent agents has to be:
• scalable: the monitoring approach is scalable
because it is based on a multi-agent system which scales well with the
growing
size of the monitored network. For that, one has to integrate an agent
(or a
group of agents) on the new node to be controlled and the monitoring of
this
node is realized;
• distributed: each agent is responsible for a local
monitoring. There is no centralization of the information collected by
the
different agents, and the decisions the agent performs are in no way
based on
global parameters. This feature is very important as it avoids having
bottlenecks around a central monitoring entity;
• adaptive: the agent adapts its actions depending on
the monitored data and entities according to the incoming events and
the vision
of the current system state. The monitoring that could be adopted is
adaptive because
of the following: (1) the agent modifies the monitored parameters: the
agent
decides, at every moment, which parameters must be monitored and which
ones are
no longer important under the current conditions; (2) the agent adapts
the
current management mechanisms (MM) and the actions undertaken when a
certain
event occurs. The actions the monitoring process executes may become no
longer
valid and must therefore be replaced by other actions. These new
actions are
considered more suitable to the current observed state;
• cooperating: the agents are cooperating to reach a
common goal. This property is pretty important in large scale
telecommunication
networks to be sure about a common convergence of the system.
• local: the agent monitors only local parameters.
However, the agent can use information sent by its
• selective: the agent filters the received events and
reacts only to those it recognizes. Event
classification is used
to trigger the appropriate actions.
The goal of the network is to mediate communications
on behalf of the user in a proactive way. Many of our current
applications are
governed by the pull model in which the user requests information or
initiates
a communication stream (Web, e-mail etc). In this interactive mode of
operation, the user is an active part and must initiate the process of
information acquisition. We believe that a personal communication
environment should
behave according to the push model in which a source of data or the
network
infrastructure takes care of preparing communications and proposing
them to the
user. In this model, it is the source or the network infrastructure
that
proposes the information to the user in a similar way to the SIP
protocol which
invites the user to a communication session. Such a mode of operation
is at the
base of the future Embedded Internet: it takes care of preparing
communications
and proposing them to the user.
The role of the multi-agent system is to mediate
notification events according to the user preferences, discover
required
services, configure them, and compose into complex communication
applications.
When an appliance enters a given zone, it authenticates itself so that
the user
preferences and device capabilities can be retrieved. Based on this
information
and the current context, the network infrastructure discovers relevant
communication
sources, component services, and applications that interact with the
user.
Then, it dynamically binds them to form a complete communication chain.
The
composition is guided by the information on user preferences, device
capabilities, the current context, and a precise description of data
types that
can be generated on output or accepted on input by each component. The
whole
process is designed for
minimal user
intervention---we would like to automate most of the operations needed
when
appliances move and change
network connections.
Task 1:
Service and Resource Algorithms
This
task concerns the definition of algorithms
that will have to control
- the allocation of flows in the physical
network depending on user profile, SLAs (Service Level Agreements), and
virtual
networks characteristics.
- the distribution of the resources between the
virtual networks.
These algorithms have to be designed within the
piloting plane. This task will define algorithms and protocols that
will run in
the piloting system. Those algorithms will export configuration
parameters,
which will allow the control plane to modify their behaviour on the fly.
This
Task will build the piloting system using
the multi-agent environment and the Behaviours (and rule engine)
proposed in WP
1. This distributed intelligent agent platform form the basis for the
Piloting
system.
The structure of the agents was shown in Figure
3. This task will have to define the dynamic planner in details and the
Behaviours dedicated to the decision process that can be use the
intelligent
Behaviour defined in WP1. The result of the work performs in the
piloting plane
is to feed the control algorithms defined in Task 1 with the best
available
information coming from the knowledge treated by the different
Behaviours.
Based
on the results that we will be obtained
using the piloting system we will have to determine what are the best
situated
views for the different algorithms defined in Task 1 of this Work
Package.
Indeed, the situated view can be definitely different depending on the
algorithm to be fed. For example, a one hop or a two-hop or a complete
situated
view can be defined but also more complex situated views as defined in
Figure
10.
Fig. 10.
Different solutions for the situated view (entire, intermediate and
one-hop)