Chapter 9

Reliable Service Discovery in MANET

9.1 Introduction
In mobile wireless environments, agents can serve multiple functions, such as service
discovery, automatic network reconfiguration, and clustering, among others. These agents
can also enhance various services within Mobile Ad Hoc Networks (MANETs). For instance,
in certain studies, agents are employed to position checkpoints near Mobile Nodes (MNs) to
facilitate quicker recovery processes. Additionally, agents can be utilized to develop intrusion
detection systems tailored for MANETs. However, prior to the widespread commercial
implementation of such applications, it is essential to evaluate the availability, reliability, and
overall dependability of the agent system. A dependable mobile agent-based system ensures
that the agents remain accessible and reliable throughout the system's operational lifespan.
In this regard, the focus of the analysis is on service discovery within MANETs utilizing
mobile agents. When a client application seeks to access a service offered by a host or server,
a service discovery process is initiated. Although this is a fundamental application for
MANETs, the lack of centralized intelligence in the network and the limited energy resources
of the MNs present significant design and deployment challenges. Numerous solutions have
been proposed; however, only a select few have effectively tackled the uncertainties inherent
in MANETs. The application scenarios for service discovery in wireless mobile environments
are diverse:
- In MANETs, some connected hosts may possess, in addition to their ad hoc network
interface, an external connection to the Internet. This capability can be considered a service
for the participating ad hoc nodes. Through service discovery, members of the MANET can
utilize such gateway services.
- Participants in collaborative applications can use their wireless handheld devices or laptops
to discover application or game servers prior to joining a session.
A service discovery protocol is deemed well-suited for MANETs if it allows for the selection
of a suitable instance of a service.
The provision of consistent services remains essential, even amidst frequent changes in
network topology. The reliability of the protocol is crucial for identifying the presence of
quality services within the network, highlighting the need to assess the reliability of such
protocols. A higher reliability level correlates with an increased likelihood of locating
services in the network. Given that agents can better manage the uncertainties inherent in
Mobile Ad Hoc Networks (MANET), an agent-based service discovery protocol has been
selected for examination. The details of this service discovery protocol are outlined in
reference [6]. Reliability is evaluated using the Monte Carlo simulation technique developed
in earlier chapters. Subsequently, a new agent-based service discovery protocol is proposed,
designed to ensure dependable performance despite the dynamic nature of the network. The
strategies employed in the protocol's design can be readily adapted to enhance the reliability
of other agent-based applications within MANET.
9.2 Service Discovery Process
In a network, a service can be defined as any hardware or software resource available for use
by other clients. The process of service discovery involves identifying services within the
network. The following methods are utilized to discover and manage service-related
information [66]:
• Service providers disseminate service advertisements throughout the network;
• Clients send out messages across the network in search of the desired service;
• Nodes store service advertisements in their cache;
• Nodes monitor network traffic and cache relevant information regarding available services.
The first method represents passive discovery (push model), while the latter describes active
discovery (pull model). The remaining two methods are derived from the initial two
approaches. Although the push mechanism incurs significant costs in terms of network
bandwidth, particularly in the context of MANET, the pull mechanism is hindered by
suboptimal performance characterized by longer response times. Additionally, other factors
must be considered, such as the network size (number of nodes), service availability (the
frequency with which services appear and disappear), and the rate of service changes.
The service discovery protocol outlined in reference [6] serves as the foundational protocol
for this study. Initially, the reliability of the Multi-Agent System (MAS), in which agents
navigate the underlying Mobile Ad Hoc Network (MANET) to identify various services
offered by the nodes, is assessed. The algorithm described in [6] employs two distinct types
of agents: a static Stationary Agent (SA) and a mobile Travel Agent (TA). The SAs are
deployed on a per-node basis, while the TAs are dynamically assigned to gather and
disseminate service information throughout the MANET. A TA favors nodes along its path
that it has not previously visited but can reach through nodes it is already familiar with. To
facilitate this, the TA Route algorithm proposed in [6] identifies the next target migration site
for a TA. The SAs oversee the population of TAs within the network, adjusting their numbers
based on the incoming agent frequency, which refers to the number of TAs visiting a node.
Depending on the available network bandwidth, an SA may either generate or terminate a
TA; a higher bandwidth allows for a greater number of agents, which can enhance
performance and potentially increase reliability.
Both approaches indicate that if the agents possess sufficient intelligence to learn and
anticipate node and link behaviors, they can effectively navigate the dynamics and
uncertainties inherent in MANETs. Consequently, by enhancing the dependability of the
MAS, the overall reliability of the MANET is also improved.
9.3 Reliability Analysis of Service Discovery Protocol
In this study, we consider that the Multi-Agent System (MAS) at a specific time t comprises
m(t) independent agents, referred to as Travel Agents (TAs) in reference [6], which are
capable of navigating within the underlying Mobile Ad Hoc Network (MANET). The term
m(t) reflects the dynamic nature of the number of TAs, as a Self-Actor (SA) can eliminate
TAs, as noted in reference [6]. The reliability of the system S is characterized by the
probability of S remaining operational over a designated time interval, as discussed in
reference [1]. The reliability of each agent is articulated through equation 4.8.
The foundational network is represented as an undirected graph G = (V, E), as outlined in
Section 4.2.1. This network comprises N nodes, thus |V| = N, which may or may not be
interconnected through bidirectional links (e). The existence of a link e at any given moment
is determined by a specific probability. The mobility of the nodes is modeled using a Non-
Homogeneous Poisson Process (NHPP) distribution, following the nhppMANET() function
detailed in Algorithm 5.
In this context, an agent can be conceptualized as a program that transitions from one node to
another within the network, contingent upon the connectivity of the nodes, according to the
strategy defined in the TA Route Algorithm in reference [6]. Each TA maintains a record of
the nodes it has visited, referred to as the Traffic Route Map (TRM). For every visited node
MNi in the TRM, a Route Sequence Number (RSN) is assigned, along with a list of
neighboring nodes of MNi (LN) that are also documented. The RSN serves to enumerate the
visits.
The prior visits of the TAs are recorded, allowing the TA to be aware of the nodes it has
recently accessed, thanks to the assistance of the RSN. The LN of a node comprises all nodes
that are directly reachable and have not yet been visited by the TA. When a TA explores a
new node, the identifier of that node is eliminated from all current LNs within the TRM.
Consequently, the LN serves as a tool for identifying nodes that have unvisited neighbors as
well as shared neighbors among nodes. An illustration of TA migration is provided.
Figure 9.1: (a)Example of Service Discovery Agents [6]; (b)Table showing TRM of the TA
under consideration shown in Figure 9.1. As shown in Figure 9.1(a) a TA, first visits MN1
and then MN4. If the TA is then moving to MN2 it contains a TRM as depicted in Figure
9.1(b). The RSNs show the order in which the TA has visited the nodes. The LNs contain all
nodes that a node can directly reach except those which are already visited. For example
MN4 is a relay node that can communicate with MN2, MN3, MN1, MN5 and MN6. Because
MN1, MN2 and MN4 are already visited they are deleted in the LN of all nodes in the TRM.
In the referenced protocol [6], the aspect of node mobility during the agents' journey is not
addressed. Consequently, essential modifications are implemented to enhance the service
discovery process, making it more compatible with the dynamic nature of Mobile Ad Hoc
Networks (MANET). A Task Agent (TA) initiates its journey from an owner, where it is
generated by a Service Agent (SA), and traverses from one node to another in accordance
with the TA routing algorithm [6]. The success of this movement is contingent upon the
connectivity between the two nodes and the absence of transient errors. Therefore, a
probability is assigned to the movement to reflect the transient characteristics of the
environment. This consideration arises from the possibility that the routing table may not be
accurately updated or that the quality of the link may have deteriorated significantly due to
increased noise levels, preventing the agents from migrating. Thus, if an agent located at
Mobile Node A (MNA) opts to transition to Mobile Node B (MNB), which is connected to
MNA, the agent will successfully move to MNB with a probability denoted as pt. Should an
agent encounter a situation where all unvisited nodes are unreachable, it will pause and
subsequently attempt the move again. This approach accommodates transient faults, such as
temporary link failures, as the agent retries after a brief delay, ultimately enhancing the
overall system [Link] is not considered in [6] but to make the service discovery
process more suitable to MANET dynamicity, transient fault tolerance becomes a necessity.
The algorithm followed by the TAs, TARouteMod() is listed in Algorithm 22. As
The agent's choice of the next destination is influenced by the currently accessible set of
nodes. It is assumed that, over short time intervals, the network topology will not undergo
significant changes that would result in the complete disconnection of common neighboring
nodes or reduce them to a node with very few connections. For instance, a TA may move to a
node after executing step 6 of TARouteMode() (Algorithm 22), only to discover that all
unvisited neighbors of that node have become unreachable.
9.3.1 Reliability Analysis with MANET Following NHPP Model
The reliability of the Multi-Agent System (MAS), which comprises the TAs, is evaluated in
relation to the network's status and conditions, such as link connectivity and path loss
probability. Each agent is anticipated to visit all operational nodes within the Mobile Ad Hoc
Network (MANET) to gather and disseminate service information. As previously noted,
transient failures are also taken into account. For simulation purposes, we have modeled the
failure probability (P) of the mobile nodes (PNode) as a variable following a Weibull
distribution. The Monte Carlo simulation algorithm, ServiceReliability(), designed to
estimate agent reliability, is detailed in Algorithm 23.
9.3.2 Reliability Analysis of MANET with Mobility Models
Modeling the links in a Mobile Ad Hoc Network (MANET) using Non-Homogeneous
Poisson Processes (NHPP) does not adequately account for relative mobility. In practice, the
relative speed of nodes primarily determines the existence of links. Therefore, mobility
models can be employed to simulate node movement across various application scenarios.
Numerous mobility models have been proposed to tackle this challenge, tailored to specific
contexts such as disaster response or military operations. The following three mobility
models are examined:
• Random Waypoint Mobility Model [36]
• Smooth Random Mobility Model [100]
• Reference Point Group Mobility Model [130]
[Link] Random Waypoint Mobility Model (RWMM)
The Random Waypoint Mobility Model (RWMM) [36] is frequently utilized to simulate node
mobility due to its straightforward nature. In this model, a mobile node randomly selects a
destination point (waypoint) within the designated area and travels in a straight line to that
point at a constant speed. After a predetermined pause duration, the node selects a new
destination and speed, then proceeds to the new waypoint at a constant velocity. In this
context, we have defined a linear velocity vi(t) and a direction φi(t) to calculate the
subsequent destination point.
However, RWMM may lead to abrupt changes in direction or sudden halts when the
differential rate of velocity change approaches infinity, which is impractical in real-world
situations [100].
[Link] Smooth Random Mobility Model (SRMM)
The application of SRMM in modeling MANET has been thoroughly examined in Section
4.2.2. This model seeks to incorporate temporal dependency, whereby the speed and/or
direction of a mobile node (MN) at a given moment is influenced by its previous speed and
direction. While this approach is quite realistic, certain scenarios, such as movements on a
campus or during disaster relief efforts, typically involve users moving in groups.
Consequently, the movement of one individual can impact others, a factor that SRMM does
not account for.
[Link] Reference Point Group Mobility Model (RPGM)
In contexts such as battlefields or rescue operations, individuals often operate in groups,
making group movement a common occurrence. Therefore, RPGM can also be employed
[130]. In this model, the speed and direction of mobile nodes (MNs) are determined by their
leader, referred to as the reference point. The leader's velocity can again be modeled using
RWMM. If v_leader and φ_leader denote the speed and direction of the reference point,
respectively, then the speed (v_i) and direction (φ_i) of the mobile node (MNi) can be
computed as follows [130].

In this context, SDR and ADR represent the speed and angle deviation ratios, respectively,
with the condition that 0 < SDR, ADR < 1. Consequently, the position of MNi at the time
instant (t + ∆t) is estimated using equations 9.3 and 9.4. Therefore, in the current analysis, the
movement of the nodes is simulated using one of the three aforementioned methods. The
power of the received signal is computed based on the two-ray propagation model of radio
signals, as elaborated in Section 4.2.2.
[Link] Steps for Reliability Analysis With mobility models and two ray propagation model in
place, MANET can be formed. To assess agent reliability equations 4.8 and 4.9 are used.
However the performance of the service discovery protocol is measured by a node’s success
in discovering service providers in the network. All or a few nodes in MANET may act as
service providers, thus total number of service providers, SPtotal≤N. The instantaneous
reliability of service discovery process spawned at MNi (Ri(t)) can be given as

Here SPdis i denotes the number of service providers discovered by MNi . If Ri(t) is
integrated over time, it gives the overall service discovery reliability R i service as follows

Time is segmented into discrete intervals, during which snapshots of the system are captured
at brief periodic intervals. As this interval approaches zero, a nearly continuous
representation of the system becomes available. This process is repeated Q times in
accordance with the Monte Carlo simulation methodology [102]. The algorithm for
estimating agent reliability, referred to as ExtendedServiceReliability(), is detailed in
Algorithm 24. However, these algorithms do not address various challenges associated with
Mobile Ad Hoc Networks (MANET), and as a result, they may not consistently guarantee
service reliability.
9.4 A Reliable Service Discovery Protocol for MANET
This section presents a service discovery protocol designed to maximize the reliability of
Multi-Agent Systems (MAS) within a specified MANET configuration. Reliability is
assessed using the Monte Carlo simulation approach established in [4], which takes into
account the Quality of Service (QoS) requirements related to link capacity for the agents.
Each mobile agent may possess distinct QoS requirements, focusing solely on link capacity
[4]. Therefore, if a viable path exists between the agent's current location and its next
destination that meets its QoS criteria, and both nodes along this path are functional, the
agent will proceed to its next destination via that route. The required link capacity can be
modeled using a Normal distribution, as indicated in [4]. Consequently, each agent, with its
unique QoS demands, will perceive the network differently.
In this scenario, M agents originating from G groups commence their tasks from G nodes,
which serve as their respective owners. Each owner allocates the responsibility of service
discovery equally among its agents. Consequently, each agent receives a list of nodes from
which the owner requires information. The agent may select its subsequent destination at
random from the connected nodes on its list. The nodes share their service information with
the visiting agents, which can occur through message passing or via a stationary agent. The
agents, in turn, relay the information they have gathered to the nodes they visit. As a result, a
node may become aware of a service offered by another node and communicate this
information to the visiting agents. Therefore, an agent may acquire knowledge about a
service provider without directly visiting that node and can disseminate this information to
others.
Each service record is assigned a timestamp to accurately forecast the availability of services.
When an agent visits a node and discovers more recent service information regarding a
service provider, the agent's database is updated accordingly. Thus, if an agent visits a service
provider node, it refreshes its list with the timestamp indicating the duration of service
availability. However, if an agent learns about a service provider from one of the nodes it has
visited, it can only make predictions regarding that service. Consequently, indirect
information should be assigned less significance than direct observations, as predictions
about a node are more reliable when an agent has physically visited that node.
Furthermore, to prevent an overload of service discovery agents within the network, each
node has a predetermined agent tolerance limit. If this limit is surpassed, the node begins to
terminate its visitors, meaning the agents that come to visit. As a result, an agent dispatched
by the owner may not always return, prompting the use of a timer to monitor such
occurrences. When a timeout event arises, the binary exponential backoff algorithm is
employed by the owner node to generate new agents.
From equation 4.7, it is established that the reliability of the multi-agent system is contingent
upon the reliability factor.

taken the decay function, decay(t) as shown below

Thus if a node is found to provide service for a longer time, then chances are that the service
is more stable and hence the owner (and the nodes which are informed about this service by
the agent) may get it if needed. Thus λi is much influenced by the conditions of underlying
MANET like node mobility, interference, available link capacity etc. Now, the probability
that the MAS is operational can be calculated as the mean of reliability of all its components
(which are agents) in this system as follows

This equation is derived from equation 4.9. In this context, m(t) denotes the number of agents
present in the system at time t, indicating that it is not a constant (M in equation 4.9). In the
equation above, ri(t) signifies the probability that agent i remains operational during time t.
The effectiveness of the service discovery protocol is evaluated based on a node's ability to
successfully identify service providers within the network. It is possible for all or only a
subset of nodes in a Mobile Ad Hoc Network (MANET) to function as service providers, thus
SPtotal is less than or equal to N. The instantaneous reliability, denoted as Ri(t), can be
computed using equation 9.7.
By integrating Ri(t) over time and repeating this process Q times, one can derive the overall
service discovery reliability, R i service, as outlined in equation 9.8. The methodology for
discovering and disseminating service information among nodes in a MANET through
mobile agent groups is detailed in Algorithm 25 (ServiceDiscoveryAgent()). The reliability
assessment of the nodes and the updating of agent feedback are presented in
ServiceDiscovery(), as shown in Algorithm 26. Furthermore, the reliability of the agent
system is calculated using equation 9.11, while equations 9.9 and 9.10 are employed to
determine the reliability of individual agents.
It is noteworthy that a service provider node, even in the absence of spawning agents, can
gather a significant amount of information regarding other providers from its visitors.
Additionally, as indicated by equation 9.10, an agent that identifies a stable service is deemed
more reliable than one that locates a scarce but unstable service.

9.5 Experimental Results

This section adheres to the conventions established in earlier chapters, with the simulation
program implemented in Java. The initial positions of the agents are predetermined, with all
agents from the same group commencing their journey from a designated node, identified as
the owner. Initially, the experimental results pertaining to the reliability of the service
discovery process, as discussed in [6] (refer to Section 9.3), are presented. This is succeeded
by the findings outlined in Section 9.4. The results illustrate the reliability of the proposed
service discovery protocol within a specified initial configuration of the Mobile Ad Hoc
Network (MANET).
9.5.1 Service Discovery with MANET Modeled by NHPP
The default parameter values are detailed in Table 9.1. Unless indicated otherwise, these
parameters will consistently utilize the default settings. To facilitate a comprehensive
understanding of the algorithms TARouteMod() (Algorithm 22) and ServiceReliability()
(Algorithm 23), an example is provided based on the data generated by our simulation
program. In this example, a network is formed by five nodes. Every three seconds (∆t = 3),
the positions of the nodes and, consequently, the network connectivity graph are updated in
accordance with the Non-Homogeneous Poisson Process (NHPP). Five mobile agents are
deployed by the five distinct owners (nodes), each commencing their journey from their
respective owners. Thus, agents 0, 1, 2, 3, and 4 initiate their travels from nodes MN0, MN1,
MN2, MN3, and MN4, respectively, as they navigate the network to fulfill their tasks. Our
objective is to determine the number of nodes that are successfully visited.
The agents indicate the number of services they discover and the extent to which they
disseminate this service information within the Mobile Ad Hoc Network (MANET), which in
turn allows for the calculation of the agents' reliability. The average reliability of all agents
over a specified time frame across multiple simulations reflects the overall reliability of the
Multi-Agent System (MAS), despite the inherent uncertainties present in the MANET. Our
migration policy, derived from reference [6] and detailed in Algorithm 22, acknowledges that
not all reachable destinations are equally probable. As illustrated in Figure 9.2(a), at the time
t=t0, agents 0, 1, and 3 remain stationary at their respective owners, while agents 2 and 4
proceed to their neighboring nodes. Consequently, after a time interval of ∆t, agents 2 and 4
are located at MN4 and MN2, respectively. In the subsequent time step, agents 0, 1, and 2
identify services offered by MN1 and MN0 (as shown in Figure 9.2(b)). However, agents 2
and 4 do not migrate further, having already visited MN2 and MN4, respectively. In the
following time step, the connectivity graph undergoes significant changes (depicted in Figure
9.2(c)), allowing the agents to discover additional services. As the simulation progresses, the
nodes become more sparse, resulting in only agent 3 successfully migrating. By the
conclusion of the simulation, agent 3 is found to have visited four nodes, thereby
accomplishing 80% of its task, while the other agents have only completed 60%.
Consequently, the overall reliability is calculated as [3/5 × 4 + 4/5] / 5 = 0.64. This indicates
that in a MANET with a link failure probability of 0.4, where nodes can accommodate a
maximum of five service discovery agents within a designated time frame, an agent is able to
cover, on average, 64% of the network. In practice, the dynamic nature of the network
influences agent migration, leading to the conclusion that the reliability of the MAS is
contingent upon its size, particularly in larger MANETs, as demonstrated in Figure 9.3(a).
The results are derived from two distinct scenarios, one of which allows each node to receive
a maximum of 20 agents (maximum tolerated frequency).

Figure 9.2: MANET configuration along with agent migration route (a)At time t=t0; (b)At
time t=t0+∆t; (c)At time t=t0+2∆t; (d)At time t=t0+3∆t
In the alternative scenario, the maximum allowable frequency is maintained at 15. Provided
that each node receives no more than 15 agents within a brief timeframe, all agents will
successfully fulfill their tasks, assuming there are no software malfunctions and the
underlying Mobile Ad Hoc Network (MANET) remains stable, as illustrated in Figure 9.3(a).
However, if the value of M is increased beyond this threshold, the reliability of the agents
diminishes, as it surpasses the maximum tolerated frequency. A higher number of agents
permitted within the network correlates with an increase in overall agent reliability, which is
reflected on the left side of the graph in Figure 9.3(a).
Each MANET is subject to bandwidth limitations, which consequently restrict the maximum
permissible value of M over a given period. It is evident that there exists a maximum
incoming agent frequency that a node can support, as depicted in Figure 9.3(b). A higher
value in this context signifies enhanced bandwidth availability from the MANET. Therefore,
it is anticipated that with increased bandwidth, the reliability of our Multi-Agent System
(MAS) improves. Nonetheless, it is noteworthy that when M equals 20, the MAS approaches
a near steady state once the incoming agent frequency exceeds 16.

Figure 9.4: (a)Reliability variation with varying number of nodes; (b)Reliability variation
with time Now if N is increased, the overall reliability does not change appreciably as long as
M (≤30) is comparable to the agent frequency (=20) supported by the nodes (Figure 9.4(a)).
Thus our approach is found to be scalable for MANETs as big as 40 nodes. But for large M
(>30), nodes may kill some agents resulting in a drop in reliability for N>32. But this result
indicates the scalability of the service discovery approach for crowded MANET as change in
network size does not appreciably affect the reliability of

The stability of the solution is also examined. As time progresses, an increasing number of
agents will be capable of withstanding temporary link failures, thereby enhancing the
likelihood of discovering and disseminating service information, which in turn improves
reliability. However, the graph presented in Figure 9.4(b) reveals an optimal point (at 750
minutes and beyond) after which the system attains a steady state for our configuration, as
detailed in Table 9.1. This figure further suggests that, in the absence of software faults
among the agents, the reliability of the MAS may approach the ideal scenario (equal to 1).
The probability of link failures also impacts the reliability of the agents. As the number of
failed links increases, the network becomes fragmented, leading to a significant decline in
MAS reliability, as illustrated in Figure 9.5. It is evident that when the link failure probability
exceeds 0.5, the network may divide into multiple components, preventing some agents from
accessing nodes located in other parts of the network.
Link failures occur not only due to increased background noise. In fact, the primary cause of
link failures is often the rise in relative mobility of the nodes. This node mobility can be more
accurately simulated using mobility models, as demonstrated in the results presented in the
following section.
9.5.2 Service Discovery in MANET - Modeled with Mobility Models
This section illustrates the impact of mobility models on agent reliability. Other factors
affecting the reliability of service discovery agents have been analyzed in the preceding
section. The default parameter values are provided in Table 9.2. Unless stated otherwise,
The parameters consistently assume these default values. A comprehensive analysis of the
simulation outcomes is presented. The movement patterns of the nodes significantly influence
various applications within Mobile Ad Hoc Networks (MANETs), as they impact the
probability of link existence. In smaller MANETs, nodes that move randomly according to
the Random Waypoint Mobility Model (RWMM) demonstrate superior Multi-Agent System
(MAS) reliability compared to the Simple Random Mobility Model (SRMM) or the Random
Path Generation Model (RPGM), as illustrated in Figure 9.6. This phenomenon may be
attributed to density waves, where the average number of neighboring nodes for a specific
node fluctuates periodically over time. However, as the size of the MANET increases, the
performance of the MAS becomes largely unaffected by the various mobility models and,
consequently, the movement patterns of the nodes. This finding indicates that reliability
predictions made in one context, such as a campus network utilizing the SRMM, are highly
relevant to multiple other contexts, including disaster relief scenarios employing the RPGM.
The reliability of the service discovery protocol is evaluated based on equations 9.7 and .
9.8 : The evaluation of performance is conducted from the perspective of MN24. Initially,
performance is assessed based on the total number of service providers (SPtotal). An increase
in SPtotal correlates with enhanced service availability, leading to a stabilization of protocol
reliability following initial disturbances, as illustrated in Figure 9.7(a). However, the impact
of service availability is particularly significant in Mobile Ad Hoc Networks (MANETs) that
experience greater transient disruptions. As environmental conditions become more stable,
agents demonstrate improved reliability in identifying nearly all service providers, even in
situations where they are limited.
Additionally, the influence of MANET size on service discovery performance is examined, as
depicted in Figure 9.7(b). In scenarios where the MANET becomes more populated (N>30),
redundant pathways to access service providers begin to appear, even in cases of scarcity. For
this analysis, the total number of service providers is considered to be 13. The observed
decline in reliability for LFP=0.5 when the node count ranges from 25 to 30 is attributed to
the fact that the introduction of new nodes expands the MANET boundary rather than
increasing congestion. Consequently, the overall connectivity of the network deteriorates.
Nevertheless, in a stable environment characterized by low transient errors, the service
discovery protocol is capable of identifying the majority of service providers.
Furthermore, the performance of the service discovery protocol can be enhanced if nodes
exchange information regarding service providers with the agents they encounter. This
assertion is supported in the subsequent section, which discusses the protocol outlined in
Section 9.4.
9.5.3 Analyzing Performance of Proposed Service Discovery Protocol We take the help of an
example and show the data generated by the simulation program and then the detailed
analysis of simulation results is presented. We have taken an

An example is presented involving a network comprising six nodes and three agents. The
connectivity graph, along with the agents and their movements, is illustrated in Figure 9.8.
Due to the smooth transitions of the nodes, as dictated by the Smooth Random Movement
Model (SRMM), no significant alterations are observed in the connectivity graph over
successive time intervals. The connectivity among the nodes is determined based on the
required link capacity of each agent, which is modeled using a Normal distribution with a
mean of 3 and a standard deviation of 0.1. Consequently, the varying capacity demands of the
agents lead to different perspectives of the same Mobile Ad Hoc Network (MANET)
regarding node connectivity. The links that provide adequate capacity to meet the demands of
all three agents are depicted with bold blue lines in Figure 9.8, such as the link between MN5
and MN6 in Figure 9.8(a). In contrast, the thin black lines in Figure 9.8 indicate links that
satisfy the capacity requirements of only some agents. The agents are represented by callouts,
with dotted callouts indicating their previous positions. Nodes MN2 and MN4 are designated
as service providers, with MN2 also deploying agents to facilitate service discovery.
Following the initial time interval, agent 1 moves from MN1 to locate a service provider
(MN2), while agent 2 identifies another service provider node (MN4), as depicted in Figure
9.8(a). In the subsequent time intervals, illustrated in Figure 9.8(b), agent 1 disseminates the
information it has gathered to MN3. As a result, MN3 (and MN5) becomes aware of the
services offered by MN2 (and MN4) through agent 1 (and agent 2). In the following time
period, the agents continue to migrate.

The dissemination of service information within a Mobile Ad Hoc Network (MANET) is

further explored. Notably, node MN6 becomes aware of both service providers as agents 1
and 2 visit this node. The later visiting agent (agent 1) may possess knowledge about the
earlier agent’s (agent 2) information, as indicated in MN6’s database. Following this, at the
third time instance, agent 3 successfully identifies a service provider node (MN4). This
process continues, allowing certain nodes, which do not generate agents, to learn about
service providers through interactions with informed agents.
Given that three groups of agents are considered, each containing one agent, it is essential for
each agent to visit all other nodes within the network. However, as the size of the groups
increases, each agent is required to cover a reduced number of nodes. Consequently, the
reliability of the agents is anticipated to improve.
A series of experiments has been conducted to validate the proposed protocol, with the
default parameter values detailed in Table 9.3. Any modifications to these values will be
explicitly noted. Initially, the reliability of agents is assessed with increasing M for group
sizes of G=5 and G=10, as illustrated in Figure 9.9(a). A node forms a group of agents to
identify services within the network, where each agent in a group shares the same Quality of
Service (QoS) requirements. Therefore, there can be a maximum of G distinct types of agents
based on QoS criteria. It is observed that reliability improves with larger group sizes,
ultimately stabilizing. For G=5, the optimal group size is determined to be five agents per
group (M=25 in Figure).

9.9(a) The reliability of agents ultimately stabilizes, regardless of the number of agent groups
involved. Consequently, a node does not need to generate a large number of service discovery
agents that would consume bandwidth; instead, it can acquire information about service
providers from other agents.
Collaboration among agents is particularly vital in scenarios where services are limited. In
larger Mobile Ad Hoc Networks (MANETs), the task for agents becomes more challenging,
leading to reduced reliability, especially with smaller group sizes such as G=5. However, the
situation improves as the number of service providers increases, as illustrated by the gradual
rise in Figure 9.9(b), independent of the size of the MANET. Performance significantly
enhances (SPtotal=6, N=35 in Figure 9.9(b)) when at least some agents can easily access the
providers. Scalability is evident, as increasing N does not alter the overall characteristics of
the results depicted in Figure 9.9(b).
The observation that Multi-Agent Systems (MAS) with various types of agents eventually
attain a steady state is further validated in the subsequent experiment, which examines the
varying capacity demands of the agents. This is achieved by adjusting the mean and standard
deviation (SD) of a Normal distribution to generate the demanded capacity within the
ServiceDiscovery() algorithm (as detailed in Algorithm 26). The mean represents the average
capacity demand of the agent groups, while the SD reflects the variability in Quality of
Service (QoS) demands among the groups. As the SD increases for a fixed mean, the
heterogeneity among the groups intensifies. Nevertheless, this does not impede the MAS's
reliability in achieving a steady state, as demonstrated in Figure 9.10(a). Conversely, when
the QoS requirements of the groups are similar (small SD), the MAS performs more
effectively if the average QoS demands of the agents are lower. The performance of the
protocol, as described by equations 9.7 and 9.8, in relation to capacity demands is illustrated.

Figure 9.11: Performance of service discovery protocol with increasing SPs Discovering a
service becomes expensive if it is scarce and hence performance of the service discovery then
depends largely on relative position of the service provider with respect to the initiating
node(that spawned service discovery agents) or the number of neighbors (of the initiating
node) discovering the service and how well the initiating node is connected to its neighbors in
terms of link capacity (Figure 9.11). For smaller values of SPtotal the same argument applies.
But as SPtotal becomes comparable to N, the situation improves and performance at a node
(MN1 here) reaches almost steady state. For larger MANETs scarce services can become
highly expensive. In such scenario the information exchange by a node with its visiting agent
can play a key role.
9.6 Conclusion
The fundamental concept of estimating reliability in agent-based distributed applications
within Mobile Ad Hoc Networks (MANET) is applied to the service discovery process. In
this context, agents are utilized to gather and disseminate service information throughout the
network. The reliability of this process is significantly influenced by the dynamics of
MANET, particularly regarding transient link failures and the available network bandwidth.
However, variations in the movement patterns of the nodes have a minimal impact on the
reliability of the Multi-Agent System (MAS). Consequently, conclusions drawn from one
specific scenario, represented by a particular mobility model, can be applicable to other
scenarios (different mobility models) as long as the transient failure probability remains
below 0.3. The protocol's performance, evaluated through R service i, demonstrates how
agents facilitate service discovery, effectively mitigating the challenges posed by poor
network connectivity. Furthermore, a robust service discovery protocol has been developed
utilizing mobile agents capable of withstanding various uncertainties inherent in MANET. A
node designated as the owner generates multiple agents with similar Quality of Service (QoS)
requirements, assigning each a distinct list of nodes to visit, thereby ensuring comprehensive
network coverage. The agents exchange the information they gather with the nodes they
encounter while also acquiring previously unknown data from them. An information decay
function is implemented for data collected indirectly by the agents, enhancing their reliability
when a stable service is identified. The protocol has been validated, with results presented in
Section 9.5.3, indicating that it achieves a satisfactory level of reliability even in scenarios
where services are limited. As nodes share information with visiting agents, a consistent
knowledge base can be established more rapidly across the network. The agents have
demonstrated reliable performance and the ability to reach a steady state, even when
configured with varying QoS requirements.