Stefan Bosse - Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

Chapter 4

Distributed Sensor Networks

Deployment of Multi-Agent Systems in Distributed Sensor Networks

Domains and Networks 128

The Sensor Node 128

The Sensor Network 129

Further Reading 143

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

128

The growing complexity of computer networks and their heterogeneous

composition of devices ranging from high-resources server to low-resource

mobile devices demands for unique and standardized new data processing

paradigms and methodologies. The Internet-of-Things is one major example

rising in the past decade, strongly correlated with Cloud Computing and Big

Data concepts.

4.1 Domains and Networks

The sensor network is partitioned into domains. Each network node creates

a local domain that can bind multiple communication end-points. Multiple

network nodes can be grouped in domains. Single nodes can act as bridges

between domains. The network can be considered as being a global domain

with independent sub-domains. Networks can be hierarchical, creating group-

and graph-network domains.

4.2 The Sensor Node

All services of a sensor node are provided by stationary non-mobile service

agents, shown in Figure 4.1. There is a root agent, the node service agent,

which is responsible to set up and start other service agents like the event

manager and the sensor sampler agent.

Fig. 4.1 Stationary agents (green( on sensor node level providing node OS services

interacting with mobile agents (yellow) using the tuple space data base

Tuple Space

Database

Node Service

Sensor

Sampler

Event

Manager

Energy

Manager

Resource

Manager

Data

Distribution

Feature

Recognition

Data

Distribution

Feature

Recognition

Notification

(SENS,0,1024)

(SENS,2,599)

(MARK,567)

(NOTIFY,-2,2)

Operating

Services

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 129

The node service agent monitors the other node service agents. If they are

in an unresponsive or failure state, the node service agent restarts the service

agents to ensure node services availability.

4.3 The Sensor Network

A distributed sensor network as the central part of a sensing system con-

sists of multiple active sensor nodes, a set N={n

, n

, ..}, connected and

arranged in a network forming a graph G(N,C) with edges C, a set C={c

, c

..}, connecting nodes and providing communication (and eventually energy

transfer) between nodes. Each sensor node provides signal and data process-

ing for a set of sensors S={s

, s

, ..}, communication, some kind of energy

supply and management, and sensor interfaces. The connectivity of the sen-

sor nodes can be used for data and power exchange, too. Each sensor node

should provide a minimal degree of autonomy and independence from other

nodes in the network.

From the computer science point of view a sensor network can be com-

posed of a set of processes P={p

, p

, ..} performing parallel data processing

and interaction using messages M={m

, m

,..}. The processes communicate

by exchanging data by using messages, and messages are exchanges by the

nodes hosting the processes by using communication links. Message-based

communication requires protocols for synchronization and message passing.

Thus, in this sense and assuming the assumption of the above definition a dis-

tributed sensor network can be considered as being a distributed computer

performing distributed data processing. Interaction between nodes is

required to manage and distribute data and to compute information.

The properties and behaviour of a distributed system can be summarized

as follows (see Figure 4.1):

1. Processing elements (PE) represent the physical resources and the

nodes of the sensor network;

2. Processes are logical resources that perform the data processing for a

specific task;

3. Cooperative communication and inter-process communication han-

dled by message passing (instead of a centralized master-slave model);

4. Communication links (the interconnection network) are physical

resources and connects PEs;

5. Processes are cooperative by interaction;

6. Communication delay does not affect the overall distributed program

behaviour (immutability);

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

130

7. Robustness in the presence of resource failures (failures of single logi-

cal or physical resource do not affect overall system behaviour);

8. Run-time reconfiguration: adaptation of single resources or the whole

system in the presence of resource failures.

Of these physical and logical elements, specifically 1, 2 and 4 are prone to fail.

4.3.1 Communication and Network Topologies

Communication is a central part of distributed data processing in sensor

networks, influencing performance and operation significantly. Figure 4.2

(top) gives an overview of different network topologies providing connectivity

between sensor or computational nodes, covering heterogeneous networks

with computation platforms differing in computational and storage resources

(including generic computers), too.

A connection between two nodes is usually a bidirectional (serial) link,

which is capable to transfer data encapsulated in messages. Message passing

is the preferred communication method for large scale networks compared

with switched networks requiring crossbar or multi-stage Banyan (butterfly)

switches. Path finding and routing, the process of forwarding a message in the

network with the goal to deliver messages from the sender to the receiver is

required in message passing networks, and for economic reasons and the

sake of simplicity there are usually no dedicated routers in sensor networks.

Thus, a sensor or computational node is a service end-point and a message

router, too. Traditional computer networks use unique node identification

that ate unsuitable for (especially loosely or ad-hoc coupled) sensor networks.

Path finding algorithms can be classified in those using routing tables man-

aged by each network node and storing information about the network

topology (usually limited to the neighbourhood of the respective node), and

those not relying on routing tables. Routing tables provide information for

path planning in advance. Managing routing tables can be critical in resource

constrained and real-time systems, because they allocate a fairly high amount

of local storage and require computational and communication activities,

which should be minimized in sensor networks. Routing tables can only reflect

an outdated view of the network (the network configuration within a certain

time interval) and hence dealing with network changes is complicated by

using routing tables.

Communication links can be classified fundamentally in wired and wireless

technologies, and networks themselves in message passing (static) and

switched (dynamic) networks with regular and irregular topologies of a spe-

cific dimensionality (see Figure 4.2, bottom).

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 131

Fig. 4.2 (Top) Network topologies and connectivity in DSN, (Bottom) Summary of net-

work topology and technology classification

Star 1:N Topology Regular 2D P2P Mesh Topology

Mixed P2P Ad-Hoc

3D Mesh Topology

Irregular 2D P2P Mesh Topology

(a)

(b)

(c)

(d)

Topology

IrregularRegular

Technology

WirelessWired

electrical optical radio

Message

Passing

Switched

One-

dimensional

Two-

dimensional

N-

dimensional

Single

stage

Multi

stage

Crossbar

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

132

There are hybrid approaches in wireless networks. Though macro-scale

sensor networks profit from wireless technologies due to reduced installation

and deployment complexity, micro-scale network embedded in material

structure can profit from wired technologies, as these are in general more

energy efficient than their wireless counterparts (directed energy flow in the

former in contrast to non- or semi-directed energy flow in the latter).

One of the simplest and still widely used topologies are master-slave 1:N

star networks. These are mainly used in centralized processing architectures.

There is only one connection from each client to a server node (connectivity

degree 1, distance 1). The failure of the server is critical. A different situation is

shown in the bottom row (b) using a broadcast medium (bus, Ethernet) which

connects N individual nodes, with a connectivity degree (N-1) and a maximal

distance of 1. Here a failure of the broadcast medium (the interconnection) is

critical.

Two-dimensional mesh-like networks, either fully or partially connected

and using point-to-point links, are the preferred topologies for wired planar

material-integrated sensor networks having the lowest interconnect complex-

ity but still providing robustness in the presence of missing or defective links

due to path redundancy (there is more than one possible path from a source

to a destination node). They have a connectivity degree of 4 and a maximal

distance of log

N. Furthermore, the logical network topology corresponds to

the geometrical placement order of nodes.

In fully connected networks (a) each node is connected with each other

node, leading to a connectivity degree of N-1 and a distance of 1 with the high-

est degree of robustness and lowest message passing latency, but requiring

the highest resource demands. Cube networks (three-dimensional topology,

middle row, right side) provide a good compromise between the afore men-

tioned networks, having a connectivity degree of 6 and a distance of log

(resulting in a lower message passing latency compared with two-dimensional

networks). They are a special case of a generic hypercube networks (n-dimen-

sional), with very limited practical use in material-integrated sensor network

due to the complex interconnect structure.

In chain or ring networks (c) the connectivity is 2, the maximal distance N-1,

providing no (chain) or low (ring) robustness. Here, failure of a single node can

already be critical: The chain is only as strong as its weakest link.

Hierarchical networks (d) provide node partitioning in spatial bounded sub-

networks connected with each other by using dedicated routers (for example,

sub-star networks arranged in chain networks applied in [GHE10]).

A chain network has no redundancy, in contrast to two- and three-dimen-

sional mesh and cube networks (with four and six neighbour node

connections, respectively). They provide an increasing number of alternative

paths without introducing additional high connectivity complexity and costs

(number of connections are in the order of nodes).

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 133

Fig. 4.3 Abstraction of distributed computing in sensor networks with processes and

message passing

Distributed computing, i.e., in sensor networks, can be mapped on a com-

munication processes model as outlined in Figure 4.3. Communication

between processes is performed via network (or shared memory) message

passing queues.

4.3.2 Message-passing and Routing

The possible number of paths P connecting the (upper) left and (lower) right

nodes and the connectivity costs C (number of all links in a regular configura-

tion) of the network depend on the number of nodes (n,m,o) in each

dimension respectively and can be calculated from Equation 4.1 (assuming a

connecting path visits each node at most one time).

(4.1)

Figure 4.4 compares the redundancy of 1-, 2-, and 3-dimensional grid net-

works in relation to the number of nodes.

In macro-scale sensor networks nodes are often connected by using wire-

less technologies with dynamic ad-hoc network topologies. Protocol-based

routing in changing networks (regarding communication connections and

nodes) is one main challenge in the design of robust and energy-aware sensor

networks.

Pro Pro

Communication

Net.

PE PE

10101

Queue Queue Queue

Pn Pnm

Pnmo

nmo

12 3

1() (, )

()!

(, ,)

()!

!!!

()

, ,

==− = − −

=−−−

n Cnm nmmn

Cnmo mnomno

, ( , ) ,

(, ,)

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

134

Fig. 4.4 Redundancy in mesh-like networks (chain, grid, cube): possible paths from node

(0,..) to (n-1,..) depending on the network size (n nodes in each dimension)

Most traditional routing protocols rely on an address space uniquely identi-

fying nodes as well as routing tables storing network information usually

limited to a bounded spatial scope. Path finding is required for the process of

forwarding of a message, and is the main goal of routing by minimizing the

number of passed routers (e.g., nodes). Commonly there is a set of different

paths connecting nodes. In the past many routing algorithms were developed

(for wireless and wired networks) finding 1. Alternative paths in the case of

missing or non-operating connectivity in parts of the networks, and by 2. Find-

ing the shortest path (concerning the distance in hop counts and the delivery

latency). One examples is junction based routing (discussed in detail in

[BAD11]). Routing in large-scale wireless sensor networks is discussed in

detail in [LI11].

Geographic routing and addressing bases on geometrical relations

between nodes, for example, relative addressing specifying the relative dis-

tance between nodes, can be used to avoid absolute unique node addressing,

which is not applicable in large scale miniaturized networks with an initially

unknown configuration, like smart dust networks [WAR01].

The scaling of networking down to material-integrated level requires simpli-

fying and robust approaches. Some of them capable for microchip level

implementations satisfying low-resource constraints are summarized below.

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 135

They all not rely on routing tables (saving memory and computation) as well

as they are able to adapt to network changes immediately.

4.3.3 Advanced -Routing with Backtracking

Basics

Traditional -distance routing can only be deployed in regular network

structures existing in mesh or grid like networks. Incomplete networks with

missing neighbour node connectivity and irregular networks with missing

nodes (disturbing the spatial regularity) require adaptive -distance routing

strategies and backtracking to overcome these irregularities, for example, iso-

lated traps, shown in Figure 4.1.

The reliable communication protocol SLIP [BOS12A] implements smart and

adaptive routing strategies suitable to be used in technical and material-inte-

grated peer-to-peer networks with a similarity of the logical and physical

network topology, that means, the neighbourhood connectivity of nodes can

be uniquely assigned to a specific geometric direction.

The protocol message format is scalable with respect to the network size

(address size class ASC, ranging from 4 to 16 bit), maximal data payload (data

size class DSC), ranging from 4 to 16 bit length) and the network topology

dimension size (address dimension class ADC, ranging from 1 to 4).

Network nodes are connected using (serial) point-to-point links, and they

are arranged along different metric axes of different geometrical dimensions:

a one-dimensional network (ADC=1) implements chains and rings, a two-

dimensional network (ADC=2) can implement mesh grids, a three-dimen-

sional (ADC=3) can implement cubes, and so on. Both incomplete (missing

links) and irregular networks (with missing nodes and links) are supported for

each dimension class, shown in Figure 4.1.

The main issue concerning message-based communication is routing and

thus addressing of nodes. Absolute and unique addressing of nodes in a high-

density sensor network is not suitable. An alternative routing strategy is delta-

distance routing, used by SLIP. A delta-distance vector  specifies the way

from the source to a destination node counting the number of node hops in

each dimension.

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

136

Fig. 4.5 Message based communication in two-dimensional networks using delta-dis-

tance vector routing. Networks with incomplete (missing links) and irregular

(missing nodes) topologies are supported by using smart routing rules.

SLIP Message Format

A message packet contains a header descriptor specifying the type of the

packet and the scalable parameters ASC, DSC, and ADC, shown in Table 4.1.

The network address dimension ADC reflects the network topology, the size

class ASC the upper size bounds of the network, and the data size class DSC

the upper bound of the data payload. There are two different main message

types: requests and replies.

A packet descriptor follows the header descriptor, containing: the actual

delta-vector , the original delta-vector 

, a preferred routing direction , an

application layer port , a backward-propagation vector , and the length of

the following data part. The total bit length of the packet header depends on

the {ASC, DSC, ADC} scalable parameter tuple setting, which optimises appli-

cation specific the overhead and energy efficiency (spatially & temporarily).

Each time a packet is forwarded (routed) in some direction, the delta-vector is

decreased (magnitude) in the respective dimension entry. For example, rout-

ing in the x-direction results in: 



1. A message has reached the

destination iff =(0,0,..) and can be delivered to the application port . There

are different smart routing rules, applied in the order showed below until the

packet can be routed (or discarded), shown in Algorithm 4.1. First the normal

XY routing is tried, where the packet is routed in each direction one after

NODE

HDT PDT DATA

ADC

TYP

DSC

ASC

ΔΔ

1 Γω

LEN

N8 N9 N11

N10N7N6

N2 N5

N4N3

Path

D=(2,-2)

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 137

another with the goal to minimize the delta count of each particular direction.

The  part of the packet descriptor selects a preferred starting direction.

Default the x-direction is selected. If this routing strategy is impossible due to

missing connectivity, the next routing strategy is tested. The packet is tried to

send in the opposite direction, which will be marked in the gamma entry 

part of the message packet descriptor. Opposite routing is used to escape

small area traps, thereby backward routing is used to escape large area traps

or to send the packet back to the source node indicating that the packet can-

not be delivered. The routing decision is based on the actual message entries

{, , } and achieves adaptive routing reflecting the actual network topology

and the path the message already had travelled, including back-end traps,

resulting in routing of alternative paths by choosing different routing

directions.

A message is only send to a neighbour node using the particular link iff the

connection to the neighbour node was negotiated and is fully operational

concerning the sending and the receiving of messages to and from the neigh-

Entry Size [bits] Description

HDT:ADC 2 Address Dimension Class

HDT:ASC 2 Address Size Class

HDT:DSC 2 Data Size Class

HDT:TYP 2 Message type = {Request,

Reply, Alive, Acknowledge}

PDT: Num(ADC)*Bits(ASC) Actual delta vector

PDT:

Num(ADC)*Bits(ASC) Original delta vector

PDT: 2*Num(ADC) Backward propagation vector

PDT: Bits(ADC) Preferred routing dimension

(x,y,..)

PDT: Bits(ASC) Application layer port

PDT:LEN Bits(DSC) Length of packet

DATA LEN*Bits(DSC) Data

Tab. 4.1 SLIP message format (HDT: header descriptor, PDT: packet descriptor,)

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

138

bour node. For this purpose, the communication controller sends periodically

ALIVE messages to all direct surrounding nodes and waits for ACKNOWLEDGE

messages send back from the neighbour node to check the state of a connec-

tion. Non-existing nodes can be detected this way, too.

The Routing Algorithm

Alg. 4.1 Functional specification of the adaptive smart routing algorithm in pseudo

notation (



: direction in which the message has arrived, moveto sends the

message in the given direction, finally applying the route function again, {}



ordered list starting with the preferred routing dimension



if contained,



raises a signal terminating a set iteration,

?: test for link to a neighbour node

in the given direction)

TypeM=Message(,

,,,,Len,Data)messagestrcuture

={NORTH,SOUTH,WEST,EAST,..) setofdirections



‐

={SOUTH,WEST,..} setof’negative’directions



={NORTH,EAST,..} setof’positive’directions

DIM={1,2,..} setofdimensionnumbers

DEF=funiconvertcoordinatetotopologydirection

DEF=fundirconverttopology

tocoordinatedirection



DEF

route

norma

l=funM,



try

{iDIM|M.

0}



doiterateoveralldimensions

Gammaignoredforbackwardtravelandifthisdirectionisreached

ifM.

>0?((i))(0)M.

M.

)then

moveto((i))

route(Mwith

=M.

‐1,(i))

routed

ifM.

<0?((‐i))M.

=0then

moveto((‐i))

route(Mwith

=M.

+1,(‐i))

routed

done

false nosuccess

withroutedtrue success

DEF

route

opposite

=funM,



try

{iDIM}



do iterateoveralldimensions

ifM.

=0

(‐i)?((‐i))then

moveto((‐i))

route(Mwith[

=M.

+1,

=‐1],(‐i))

routed

ifM.

=0

(i)?((i))then

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 139

moveto((i))

route(Mwith[

=M.

‐1,

=‐1],(i))

routed

done

false nosuccess

withroutedtrue success

DEF

route

backward

=funM,



try

{iDIM}



doiterateoveralldimensions

ifM.

=0



‐

?((‐i))then

moveto((‐i));

route(Mwith

=M.

+1,

=‐1)

routed

ifM.

=0



?((i))then

moveto((i))

route(Mwith

=M.

‐1,

=‐1)

routed

ifM.

=‐1?((‐i))then

moveto((‐i))

route(Mwith

=M.

+1,(‐i))

routed

ifM.

=1?((i))then

moveto((i))

route(Mwith

=M.

‐1,(i))

routed

done

false nosuccess

withroutedtrue success

DEF

route=funM,





:incoming‐direction‐of‐messageM

ifM.=(0,0,..)thendeliver(M.,M)

else

ifroute

normal

(M,

)then

ifroute

opposite

(M,

)then

ifroute

backward

(M,

)then

discard(M)

In a first routing attempt, the normal routing strategy is tried with the goal

to minimize the

-vector in each dimension of an ordered set of dimensions,

starting with the preferred routing dimension . If this is impossible due to a

lack of required connectivity the message is tried to send in one opposite

direction increasing the

-vector temporarily. Opposite travel is marked in the

 entry of the message. Finally, if this strategy fails also, the message is sent

backward, finally reaching the source node again if there are no other routing

alternatives found on the back path. Some care must be taken to avoid send-

ing a message back to the direction from which it originally comes from

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

140

(resulting in a ping-pong forwarding with a live lock). To summarize, this sim-

ple routing algorithm uses decision-making based on information stored in

the message, and neighbourhood connectivity retrieved by the current node.

The hardware implementation (using ConPro and standard cell ASIC synthe-

sis, details can be found in [BOS10A] and [BOS12A]) requires about 250k

equivalent gates, including 15k Flip-Flop (FF) registers, which relates to a chip

area of  2.5mm

assuming ASIC standard cell technology 0.18m. The design

is partitioned on ConPro programming level in 34 processes, communicating

by using 16 queues. There are parallel executed receiver and sender pro-

cesses for each communication port.

Robustness and Stability Analysis

A simulation of a sensor network consisting nodes arranged in a two-

dimensional matrix with 10 rows and 10 columns was performed by using a

multi-agent model. Messages and sensor nodes were modelled with agents. A

comparison of XY and smart routing using the routing rules introduced in

Algorithm 4.1 is shown in Figure 4.6. The diagram shows the analysis results of

operational paths depending on the number of link failures. A path is opera-

tional (reachable) iff a node (device under test), for example node at position

(2,2), can deliver a request message to a destination node at position (x, y)

with x2  y2, and a reply can be delivered back to the requesting node. A

failure of a specific link and node results in a broken connection between two

nodes. The right image in Figure 4.6 shows an incomplete network with 100

broken links.

Fig. 4.6 Robustness analysis with results obtained from simulation (left) and snapshot

of sensor network (right)

Request Agent

Reply Agent

Node Agent

Connection

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.3 The Sensor Network 141

Fig. 4.7 Stability analysis (live locked messages, left) and snapshot of sensor network

with trapped messages (right)

With traditional XY routing there is a strong decrease of operational paths,

from a specific node (DUT) to any other node, if the number of broken links

increases. Using smart routing increases the number of operational paths sig-

nificantly, especially for considerable damaged networks, up to 50%

compared with XY routing providing only 5% reachable paths any more.

Results from stability analysis shown in Figure 4.7 point out unstable behav-

iour under some particular network topologies. Though most situations are

live lock free, there are some live locked messages circulating for ever in some

isolated traps, shown for example in the snapshot on the right side of this

figure.

The previous introduced simple adaptive routing strategy with an optimized

scalable message header outperforms traditional table-based and diffusion-

based routing strategies in terms of resource and path-search efficiency

(details of common routing algorithms in DSN can be found in [LI11]). Live-

locks can be avoided and the visited node count can be improved by adding a

path-node list memory stored in the message, as proposed and evaluated in

[BEN13]. But the improvement in delivery probability, delivery latency, and

energy efficiency using a path history for the prediction and selection of alter-

native paths is compromised by a significantly increased size of the message

header, and additionally the header has a non-fixed size, which makes the

processing of messages more difficult.

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

142

4.3.4 From Passive Messages to Active Agents

The previously adaptive routing of messages is used in various mobile MAS

presented in this book. The routing, originally part of the network node (oper-

ating) system is shifted towards agents and part of the agent behaviour, which

has an additional routing goal, making routing active by the agent itself. Exam-

ples can be found in Chapter 9. The nodes provide direct neighbourhood

connectivity only.

Active mobile agents used for communication and data distribution in sen-

sor networks has the advantages of adaptivity, ad-hoc behaviour, and forming

of self-organizing communication structures between agents. This includes,

e.g., the election of relay nodes in domains of networks and the propagation

along dynamic travelling paths between multiple data source and sink nodes,

shown in Fig 4.8.

Fig. 4.8 The comparison of traditional server-client communication (left) and mobile

agent model (right)

The mobile agent routing is a kind of complex combinatorial optimization

problem that solves the optimal path according to a sequence of visited

nodes and addressing energy efficiency in sensor networks, especially con-

cerning wireless sensor networks (WSN). Ant colony optimization (ACO) is put

forward to solve migration of mobile agent.









S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

4.4 Further Reading 143

4.4 Further Reading

1. F. Hu and Q. Hao, Intelligent Sensor Networks, CRC Press, 2013, ISBN

9781420062212

2. M. J. McGrath and C. N. Scanaill, Sensor Technologies Healthcare, Well-

ness, and Environmental Applications, Apress Open, 2014, ISBN

9781430260134

3. R. Tynan, D. Marsh, D. O’Kane, and G. M. P. O’Hare, Intelligent agents for

wireless sensor networks, in Proceedings of the fourth international joint

conference on Autonomous agents and multiagent systems -

AAMAS’05, 2005, p. 1179.

4. A. Rogers, D. D. Corkill, and N. R. Jennings, Agent Technologies for Sensor

Networks, IEEE Intelligent Systems, vol. 24, no. 2, 2009.

5. W. Liu, Data Driven Intelligent Agent Networks for Adaptive Monitoring and

Control, Michigan Technological University, 2012.

6. S. S. Iyengar and R. R. Brooks, Distributed Sensor Networks. CRC Press,

2005.

7. Stefano Bonucelli, Agent-Based Routing Algorithms for a Wired Sensor

Network, Università degli Studi di Genova Scuola Politecnica, 2014.

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)

Chapter 4. Distributed Sensor Networks

144

S. Bosse, Unified Distributed Sensor and Environmental Information Processing with Multi-Agent Systems

epubli, ISBN 9783746752228 (2018)