Mobile Agent-based Policy Management

for Wireless Sensor Networks

ZHOU Ying Xiao Debao

Dept. of Information Engineering Dept. of Computer Science
WuHan University of Technology HuaZhong Normal University
WuHan, HuBei, China WuHan, HuBei, China

Abstract—Wireless sensor network is a revolution of information

sensing and collecting. This kind of networks is composed of
hundreds and thousands of sensor nodes in small size with
ubiquitous sensing and computing capabilities. The task of
management of Wireless sensor networks in environments with
hundreds and thousands of sensor nodes is very complex. Until
now, wireless sensor networks and their applications have been
developed but with the management research in the germinal
stages. In this paper, we describe a management architecture for
wireless sensor network that combines mobile agent with policy
technologies, and take into account the restrict physical resources
of this type of network in the form of policies, such as energy and
computing power, imbalance of network flow, and faults caused by Figure 1. Model of wireless sensor network
nodes unavailable. Then the paper proposes the hierarchical
architecture of policy management and mobile agent-based Nowadays, a wireless sensor network is becoming an
network management structure. increasingly important technology that will be used in a variety
of applications. The large use of wireless sensor networks
Keywords-Wireless sensor network; Mobile Agent; Policy; depends on the design and development of a scalable, low-cost
network management sensor network architecture. This kind of networks is highly
dynamic in nature, as sensor networks are formed over wireless
I. INTRODUCTION links. So, the task of management of wireless sensor networks in
Wireless sensor network is a revolution of information environments with hundreds and thousands of sensor nodes and
sensing and collecting. It combines sensor technology, embedded the dynamic nature of the network is very complex.
computing, distributed information processing and wireless
communication technology all together, to realize monitoring, Until now, wireless sensor networks and their applications
sensing and collecting various environmental data in the have been developed but with the management research of
monitored area. A wireless sensor network consists of hundreds wireless sensor networks in germinal stages [10]. This will
and thousands of sensor nodes, which are tiny, low-cost, definitely be a critical problem when the applications for wireless
low-power radio devices dedicated to performing certain sensor networks require a large number of sensor nodes. The
functions such as collecting data and sending them to need to coordinate such large networks as well as their inherent
infrastructure processing nodes. They buildup the wireless sensor limitations like power constraints, limited bandwidth, distributed
networks based on their collaborative efforts as shown in Fig.1 coordination and deployment lead to a number of challenges in
[3]. The sink node can be considered to be an access point, and it the design and deployment of a wireless sensor network.
communicates between task manager node with sensor nodes by In order to simultaneously satisfy these requirements and
Internet or satellite. constraints, and improving the overall efficiency, we need the
management operations of wireless sensor networks. In this
paper, we present a mobile agent-based policy network
management system for wireless sensor networks, which aims at
providing effective management mechanism in the form of
policies, in order to satisfy the networking requirements and have
them automatically realized in the network by mobile agents, and

0-7803-9335-X/05/$20.00 ©2005 IEEE 1207

keep the wireless sensor networks running in a normal, stable III. POLICY TECHNOLOGY
and reliable way with high efficiency. Policy-based networking[2], is a powerful approach to
II. MOBILE AGENT TECHNOLOGY automating network management, as evidenced by the numerous
industry efforts in this area dealing with diverse networking
In this paper, we introduce mobile agent technology into the domains, e.g., configuration management, quality of service
network management architecture. Mobile agent is a section of control [1], [9], traffic engineering [6], etc. A policy implies a
code that can implement user tasks autonomously. A mobile pre-determined action pattern that is repeated by an entity
agent can suspend its execution on a node, and then migrate and whenever certain system conditions appear [4].
transfer its code, data state, to another node on the network, and
resume execution and access the resources on the new node. More recently, in the area of distributed systems
When completing the task, the mobile agent returns the results to management, work in [8] outlined an approach to providing
the source node. Each host provides an execution environment automated support for analysis of policy hierarchies for the
for one or more agents, and the underlying physical network management of very large distributed systems. Then, the use of
provides the means for agents on different hosts to communicate policy-based management for QoS management in wireless ad
and migrate [6]. hoc networks has been studied.
Mobile agent is the combination of distributed and artificial The architectural components of the Policy Framework
intelligence technologies and it is a kind of intelligent agent with include a Policy Decision Point (PDP) and Policy Enforcement
mobility, the main concept of it is to execute in distribution. It Points (PEP), and a policy repository. The policy rules stored in
can distribute the local tasks to be executed on the nodes with the policy repository are used by the PDP, which defines the
data or nearby nodes, and finally return the executing results. In processes and results of policy rules. It interprets and translates
this way, there is no need to collect all the data executed on only the policy data of logical entities to a device-dependent format
one node. Thus, it can save the network communication and configures the relevant PEPs. The PEP enforces the logical
bandwidth effectively, and improve the degree of parallelism and entities decided by policies, including active PEPs and passive
real-time performance of the task execution. PEPs.
The characteristics of mobile agent are dynamic execution, Once policies are defined, they are automatically enforced by
asynchronous computation, parallelism, intelligent routing. In the the management system. These capabilities can provide powerful
heterogeneous network, mobile agent can migrate to and interact functions to configure and control their network, and to
with the outside autonomously. The advantages of mobile agent re-configure their network in response to network conditions,
are listed as follows: with the highest possible level of automation.
1) Reduce the flow quantity, connecting time and number of Network management functionality can be realized by policy
times in network: one the one hand, mobile agent can execute agents that are organized in a hierarchy to provide both
scalability and autonomy. Survivability is achieved by enabling
many tasks in right server autonomously and locally, so the flow
any component to take over the management role of another
quantity through the network will reduced to the lowest. On the component in the case of failure.
other hand, mobile agent can execute asynchronously, and
reduce the unnecessary network connecting time. IV. ARCHITECTURE OF WIRELESS SENSOR NETWORK
2) Integrate with distributed structure seamlessly: the MANAGEMENT
strongpoint of mobile agent lies in its supporting for distributed
A. Structural model of network management
structure. Mobile agent can be distributed in many servers,
instead of only two servers in tradition, and distribute the tasks Based upon the information collection and communication
strategy, there are three types of network management structural
properly. Mobile agent will lead the work in resource
models: centralized, distributed, and hierarchical [3]. In a
concentrated node to the node with less work. In the period of centralized network management system, there is a single
peak-load in the network, mobile agent will delay its own tasks manager station that collects information from all nodes and
or execute the work with smaller load only. controls the entire network. This is an easy implementation but
3) Pervasive computing: mobile agent technology keeps away with some potential problems. For example, the manager station
from the bottleneck of network and processing power in is a single point of failure. Also, if a network partition occurs
pervasive computing. It transfers the interaction and because of some kind of failure, the portion that is disconnected
information processing to be executed in host with strong from the manager is left without any management functionality.
processing ability and security, which is one of the pervasive In a wireless sensor network, a centralized model will suffer from
computing technologies with good applied cost. a high message overhead in data collection. A distributed
management system has multiple manager stations; each
manages a subnetwork and communicates with other manager
stations in a peer-to-peer manner. Using the distributed approach,
a network management system could achieve higher reliability

0-7803-9335-X/05/$20.00 ©2005 IEEE 1208

and efficiency as well as lower overhead both on communication locally, or domain-wide) by automatically reconfiguring the
and computation resources [7]. Thus, in a large network as needed to deal with fault and performance problems.
telecommunication network, a distributed model is a better In this architecture, any node can dynamically take over the
choice. But it is complex to realize and control. functionality of another node to ensure survivability. A flexible
agent infrastructure allows dynamic insertion of new
Hierarchical network management systems shown in figure 2 management functionality.
integrate the advantages of the first two management models [5],
and use intermediate managers to distribute the manager tasks.
Each intermediate manager has its domain called cluster; it
collects and processes node information of its domain and passes
the information to the upper level manager if necessary, and the
intermediate manager is called cluster head. It also distributes the
messages from the upper level manager to nodes in its domain.
There is no direct communication between intermediate
managers.
Figure 3. Hierarchical architecture of policy management

C. Mobile Agent-Based Network Management Structure

The policy manager and policy agents of the two levels can
all be viewed as an agent with capability of policy enforcement.
A complete node in the network includes agent execution
environment, policy agent and the real managed node. The
Figure 2. Structural Model of WSN management execution environment supplies necessary conditions for policy
enforcement and agent migration. A policy agent is the key part
However, there is also a problem in the imbalance of network for management, and the managed nodes are common sensor
flow, if this structural model above used for wireless sensor nodes. The structural of a policy agent is shown in figure4.
network management, because the information flow from to
cluster head and nodes to the manager is obviously less than that
in opposite direction. Therefore, we can apply two technologies
to the management of wireless sensor network, one is mobile
agent, and another is policy management.
B. Hierarchical Architecture of policy management for WSN Figure 4. Structure of a Policy Agent
Policy management is the supply, implementation and
verification of the policies. It allocates the resources for In the structure, the policies are rules, which define the
user-workgroups and applications. standard of resource access and usage. A policy is an aggregation
of rules for devices selection in network, and each policy rule is
The hierarchical architecture of policy management for WSN composed of a series of conditions and relevant operations. The
is shown in Figure3. As shown here, corresponding to the conditions define the validity of this rule. Once this rule is
structural model above, three levels of this hierarchical activated, the relevant operations will be executed in a specific
architecture begin from the access point. They are Policy rule. The can be defined to executed while the conditions prevail,
Manager (PM) at the highest level, Cluster Policy Agent (CPA), or not. The policy information base (PIB) is an ordered set
Local Policy Agent (LPA). representing data types, the structure of which is similar to that of
The node with best configuration and most dump energy SMI in SNMP.
usually acts as the cluster head. An LPA manages a node. LPAs In original, the wireless sensor network is self-forming,
perform local policy-controlled configuration, monitoring, clustering and self-configuration. When the user is to execute
filtering, and reporting, thus reducing management bandwidth some management function, the network manager firstly sends
overhead, discovering the topology change and improving the the relevant policy to access point. The access point puts it into
network performance. A CPA can manage multiple LPAs. A PM PIB, and sends the policy for relevant management function to
manages multiple CPAs; it is also a policy agent actually. A PM the cluster head. Then cluster head sends the policy to common
is equivalent to a global PDP, and a CPA is an intermediate PDP, nodes, and the common node informs the agent with data to the
and a LPA is a PEP. cluster head after received information. If the agent finds that is
These collections of Policy Agents manage all the nodes in not the target node, then returns to access point and be destroyed,
the wireless sensor network. Policies are disseminated from the with data backup in access point.
PM to CPAs to LPAs, or from CPAs to LPAs. Policy Agents Because each agent is kept on standby any time, so the
react to network status changes on various levels (globally, real-time performance is improved effectively, and the data

0-7803-9335-X/05/$20.00 ©2005 IEEE 1209

transmission by agents reduces the communication flow in the [1] R. Bhatia et al., “Policy Evaluation for Network Management”,
network. INFOCOM2000.
[2] R. Chadha et al., “PECAN: Policy-Enabled Configuration Across
V. SUMMARY Networks”, IEEE 4th International Workshop on Policies for Distributed
Systems and Networks, Como, Italy, June 2003.
Wireless sensor networks represent a new frontier in the [3] I.F. Akyildiz, W. Su*, Y. Sankarasubramaniam, E. Cayirci. Wireless
development of technology to be used in a variety of applications sensor networks: a survey. Computer Networks,38 (2002) 393–422.
in the future. But presently, most researches of wireless sensor [4] Linnyer Beatrys Ruiz，Jose Marcos Nogueira and Antonio A. F. Loureiro.
networks are concentrated on energy efficiency, routings, and MANNA: A Management Architecture for Wireless Sensor Networks，
locations and so on; the research on management of wireless IEEE Communications Magazine, 2003.2
sensor networks is just at starting stage. In this paper, we propose [5] Wenli Chen, Nitin Jain, Suresh Singh. ANMP--ad hoc network
a mobile agent-based policy management for wireless sensor management protocol. IEEE Journal on Selected Areas of
Communications, vol.17, no.8, 1999.8 1506-1531
network. The management system described in this paper shows
[6] R. Chadha, G. Lapiotis, S. Wright, “Policy-Based Networking”, IEEE
that a reduction in management traffic overhead can be Network special issue, March/April 2002, Vol. 16 No. 2, guest editors.
achievable by mobile agents and policy management. This [7] W. Chen, N. Jain and S. Singh, “ANMP: Ad hoc Network Management
architecture reduces the network management requirement for protocol”, IEEE Journal on Selected Areas in Communications17(8)
bandwidth, and upgrades the imitativeness, adaptability and (August 1999) 1506-1531.
reconfigurability of network management system. It has a good [8] Migas N, Buchanan WJ, McArtney K. Migration of Mobile Agents in
practical and research value for geographically distributed Ad-hoc, Wireless Networks. Proceedings of the 11th IEEE International
network environment. Conference and Workshop on the Engineering of Computer-Based
Systems (ECBS’04) 2004
The wireless sensor networks management is very complex, [9] Maxim Peysakhov, Donovan Artz, Evan Sultanik,William Regli. Network
and there is a great deal of scope for further research in this area. Awareness for Mobile Agents on Ad Hoc Networks. AAMAS'04, July
It would be interesting to explore the security communication 19-23, 2004, New York, New York, USA. 2004 ACM
between manager station and cluster heads, the setting of [10] Wang Feng, Tian Qichuan, Gao Quanxue, Pan Quan. A Study of Sensor
management functions of manager station, the survivability of Management Based on Sensor Networks. Proceedings of the 2003 IEEE
International Conference on Robotics, Intelligent Systems and Signal
the network, and so on. Processing, 2003.10

REFERENCES

0-7803-9335-X/05/$20.00 ©2005 IEEE 1210