A Cognitive Theory-based Opportunistic Resource-Pooling Scheme for Ad hoc Networks

1.1 Opportunistic Computing

OC is a type of pervasive computing and a nomenclature, and thus, it takes the benefits of randomness and uncertainty in the environment in order to establish communication between pairs of diverse devices and applications opportunistically. It exploits the transient unpredicted contacts to cooperate and share each other’s content, resources, and services [14, 28]. The opportunity comes whenever the nodes move in close contact, i.e. within the single hop, direct communication, selected intermediate, source, or destination node. Such nodes are said to be in opportunistic contact if they share a common goal, common resources, and similar context activities such as same background or affinity.

For example, let us consider a social self-organizing pervasive network having several nodes with the orchestration of resource such as processing power, memory, software application, sensors, cameras, and so on, as shown in Figure 1. In a service-oriented computing scenario where Alice is a service manager who needs to answer a number of queries from various end users, there is increased waiting time for the queries in the queue. With OC, Alice can get the queue management be taken care by trusted colleagues Bob and James, who came as her opportunistic contact during past intervals. Hence, an OC model will pool up the resources to make them seamlessly available for applications. By implementing the OC model based on cognitive theory, the beliefs [31] on the resource availability and nature of nodes can be obtained. This model improves operational robustness, as it can efficiently provide intelligent pre-failure avoidance and reduce the operational cost. The CAs can model the real world [9]. They can provide information on the availability of various resource with minimum resource management overhead. Thus, this model is proposed to be implemented using CAs.

Figure 1:

Logical architecture of opportunistic service-oriented computing scenario.

The scenario consists of orchestration of resources, where Alice’s, the service manager, task of queue management is taken up by trusted colleagues Bob and James on opportunistic contact.

1.2 Cognitive Agents

CA [12, 17] is an intelligent software agent. It is an autonomous and responsive entity that reflects the state of information of the environment, which is represented using cognitive terms. Similar to human perceptual experience and understanding of the surrounding environment, CAs can provide an immediate or controlled response to the changes in the environment. Owing to its nature of humanness, CAs can be cloned, dispatched, and activated or deactivated based on changes in the environment. An agent is created, unchanged, and executed on the agent platform. An agent platform comprises an agent server, agents, agent transport mechanism, and an agent execution environment. Usually, inter-agent communication will be done using message passing. The mobile agent code is platform independent, and it can be cloned and executed at any remote node in the heterogeneous network environment. The agent can update its information base while interacting with other agents in the environment. Thus, an agent-embedded model will be more flexible, adaptive, very intuitive, and user-friendly [10, 35]. CAs are mainly characterized by self-learning, self-improving, and self-instructing ability [20, 32], thus reducing network overhead. CAs are suitable for adaptive, dynamic, context-based behaviorism, decision making, human-mobile network interaction system [2, 26, 27], and so on.

1.3 Proposed Cognitive Agents-Based Opportunistic Scheme for Resource Pooling

The proposed opportunistic scheme works with CAs. The instance of mobile cognitive agents (MCA) runs on every registered node in ad hoc networks. For a node in need of a resource to execute a task, it should collaborate with other devices that come to its opportunistic contact and that have the required resources for which the MCA generates the belief about the resource availability, based on the observations generated by various observed behaviors on the node. It will communicate with all reachable MCAs through a request (REQ) and reply (REP) model to obtain the beliefs from the neighboring nodes. The belief record stored these beliefs and updated them periodically.

1.4 Organization of the Article

The remainder of the article is organized as follows: Section 2 gives related work on essential resource-pooling approaches in MANETs with their working principle and limitations. Section 3 provides definitions of terminology used in the proposed resource-pooling schemes. Section 4 discusses the proposed CA-based belief formulation scheme for resource pooling and its architectural functionality. Section 5 gives the performance analysis of the scheme. Section 6 presents the experimental scenario with the results. Finally, the article concludes with future work.

3.1 Behavior

Behavior quantifies the response to the occurrence of the phenomenon in the network during resource discovery. The proposed scheme will make use of a set of behavior parameters such as velocity of node, frequency of interface, resource mapping history, and so on. The parameters are to be captured within its environment or from the log to produce the behavior. These behavior parameters are quantified as low, moderate, high, sparse, dense, random, or deterministic, and so on. The number of parameters may vary from one behavior to another, and the different combination of the same set of parameters may result in distinct behaviors. For instance, BP={BH₁, BH₂, BH₃, …, BH_n} is the set of behaviors produced during a session based on the set of captured behavior parameters {bh₁, bh₂, bh₃, …, bh_n}. The probability of occurrence of an i^th behavior BH_i as an outcome of the set of captured behavior parameters {bh} is given by

P(BHi) = ∑bhk ∈ BHiαbhk × current value of (bhk)/maximum possible value of (bhk),

where α_bhk represents the weight and is given by

αbhk = ∑k = 1nbhk∑l = 1n∑m = 1nbhl, m,

where bh_k is the k^th behavior parameter captured and bh_l,m is the l number of behavior parameters each captured m number of times.

Some of the behaviors, along with their possible values, are given below:

1. Behavior velocity (high, medium, or low) is associated with the velocity of the node movement, which represents the mobility of the node.

2. Behavior node availability (sparse, moderate, or available) is associated with the contact period of the node, which represents the duration of node availability. Similarly, behavior inter-communication (sparse, moderate, or dense) is associated with the interface period.

3. Behavior location tracing (random, semi-deterministic, or deterministic) is associated with the node mobility pattern and represents the accuracy in meeting the node.

4. The behavior resource mapping history (poor, moderate, or rich) is associated with the resource availability probability of the node and represents the ability of the node to provide the resources.

3.2 Observation

Observation quantifies the act of acquisition and summarization of produced behavior. In the proposed scheme, observations such as the node is static, dynamic, highly dynamic, or high profile (resource-rich), medium profile, or low profile (resource-poor) are confirmed based on the produced behaviors, as shown in Figure 2. For instance, let OB={ob₁, ob₂, ob₃, …, ob_n} be the set of observations obtained from the set of behavior {BP}. The probability of occurrence of an observation ob_i is given by

Figure 2:

An Instance of Belief Generation Over the Resource Availability Using BOB Model.

(A) Behavior parameters captured from external environment or produced by log. (B) Set of behaviors produced based on captured parameters. (C) Set of observation summarized from behaviors. (D) Beliefs generated from the observations.

P(obi/BHj) = ∪jP(BHj) × P(BHj/obi)∑k = 1nP(BHj/obk) × P(obk),

where P(BH_j) is the probability of occurrence of a behavior BH_j∈BP.

3.3 Belief

Belief quantifies the faith (trust) on a node for resource collaboration based on the interned observations. In the proposed scheme, four types of beliefs namely patron, slack, casual and vagrant are generated, as shown in Figure 2. These beliefs indicate the nature of the node showing resource collaboration is possible or not. For instance, let BL={BL₁, BL₂, BL₃, …, BL_n} be the set of beliefs generated from the set of observations {OB}. The probability of generation of belief BL_i is given by

P(BLi/OBj) = ∪jP(OBj) × P(OBj/BLi)∑k = 1nP(OBj/BLk) × P(BLk),

where P(OB_j) is the probability of occurrence of an set of observation OB_j. Some of the beliefs and their roles in resource collaboration includes the following: (1) patron indicates that the node can actively donate the resources; (2) casual indicates that the node is uncertain in providing the resources; (3) slack indicates that the node often uses the resources for itself; (4) vagrant indicates that the node can not provide any resource support.

3.4 Belief Record

Belief record is a tuple 〈Node ID, Belief generated〉 that stores the generated beliefs. The belief generated by the MCA running on the host node will be stored as self-belief. Beliefs obtained from all the reachable neighboring nodes through the request and reply model are stored along with their node ID, as shown in Table 2.

4.1 Cognitive Agent Migration

The MCA of the navigation controller, being the registration authority, will clone itself on the newly registered device with the necessary initial handshaking. The copy of the agent from the registration authority will migrate onto the new device, ready to capture the behaviors and formulate the beliefs about resource availability. The MCA on the new device (node) will be available for communication with the MCAs of the other nodes.

The MCA on a node will communicate with reachable MCAs to collaborate with them for pooling of resources. Let us assume some of the nodes mentioned above are mobile (node belonging to rescue team and node of bystanders) and others are static (navigation controller, vehicle controller); some of the nodes are resource rich (nodes belonging to rescue team, navigation controller, and vehicle controller) and others are resource poor (nodes belonging to the bystander). To have a continuous connectivity in such a heterogeneous and volatile condition, the MCAs generate the beliefs and commute them to other MCAs.

4.2 The Cognitive Agents-Based Opportunistic Scheme for Resource Pooling

The architecture of the CAOSR, with its functional components, is shown in Figure 5. The MCA runs on the node and communicate with its reachable MCAs through the REQ and REP model. These REQ and REP mainly communicate the beliefs among the nodes. MCA is a footprint mobile agent responsible for formulating the beliefs about the availability of resources on the node. The captured behavior parameters become the input for the MCAs’ belief formulator logic. The belief formulator generates the belief about the availability of resources. The MCA stores these beliefs in the belief record along with the beliefs communicated from the neighboring nodes. It also continuously monitors the host nodes RR and proactively finds the demand for resource collaboration (prior to the actual requirement).

Figure 5:

The Architecture of CAOSR with MCA as a Main Functional Component.

(A) MCA will pass the captured behavior parameters to the belief formulator and gets back the generated belief called as self-belief. (B) The belief record stores all the collected beliefs. (C) The MCA communicates with all the reachable MCAs through the REQ and REP model.

4.3 Belief formulator

The logic, as shown in Figure 6, is based on the BOB model. The three constructs, namely behaviors identifier, observations generator, and beliefs formulator, are put together to form a belief formulator logic. The belief formulator will formulate the belief over the availability of resources on a node in an ongoing session.

Figure 6:

The Components of Belief Formulator: Three Constructs.

(A) Behavior identifier: captures the behavior parameters and produce behaviors. (B) Observation generator: summarizes the favorable behaviors into observations. (C) Belief formulator: generates the belief based on observations and stores in belief record.

The behavior identifier gets the captured behavior parameters {bh} as input and using them probabilistically produces the behaviors as shown in Algorithm 1. The probabilistic combination of the various behaviors leads to observations. These summarized observations are stored in the observation storage as in Algorithm 2, which are used to generate the beliefs.

The favorable observations are probabilistically combined to generate any of the four mentioned beliefs. These generated beliefs are stored in the belief record along with the beliefs obtained from the reachable neighboring nodes, which will be updated periodically as in Algorithm 3. The generated beliefs will provide the proactive opportunity for resource collaboration among the nodes, which can be further incorporated as a part of routing protocols, where the path establishment/re-establishment phases can be efficiently done with minimum time and link failure.

5.1 Reliability

In CAOSR, reliability (R) can be defined as the ability of the MCA on the node to provide the belief on the availability of the resource. The MCA providing the information on the availability of resources depends on the probability of occurrence of resourceful opportunistic contact between the MCAs, P_OC. The P_OC can be obtained by the number of resourceful opportunistic contacts (an opportunistic contact, able to give the resources) between the MCAs, N_Roc, and the total number of contacts, N_tc, i.e.

and the total number of contacts is the sum of the number of resource’s fewer opportunistic contacts (N_OC), the number of direct contacts (N_DC), and the number of resourceful opportunistic contacts (N_Roc) and is given by

where N_Roc is the sum of the number of homogeneous opportunistic contacts (N_hom) and heterogeneous opportunistic contacts (N_het) between the MCAs, i.e.

The higher the value of (P_OC), the more reliable is the scheme, i.e.

5.2 Convergence Rate

The convergence rate of the resources (ξ_RA) is defined as the rate of finding the resource availability and to formulate the beliefs over it. In CAOSR, a high ξ_RA is desirable, as the MCA on the node needs to be proactive in finding the resource and formulating the beliefs over it. The ξ_RA depends on the mobility of the node (N_m), which can be determined by the pause time (the node stops for a duration) (t_p), i.e.

and the node density (N_d), given by

The rate of resource convergence increases with shorter pause time (higher node mobility, i.e. nodes become dynamic) and higher node density, with which more MCAs will come into opportunistic contact.

5.3 Latency

The latency of the scheme is a function of the overall time, including the convergence time in the formulation of belief, using CAs based on BOB model for resource pooling.

5.4 Belief Formulation Time

The belief formulation time is the total time taken in the formulation of a belief, i.e. T_tot. It includes the time taken by the MCA in formulating the belief T_bf and time taken for belief exchange with the neighboring node T_bx, as shown in Figure 7:

Figure 7:

Belief Formulation Timing Diagram of MCA1 that Exchanges the Belief with MCA2.

T_tot is the total time taken in formulation of belief and is the summation of self-belief formulation time T_bf and belief exchange time T_bx. (A) T_tot is small when MCA2 formulates the self-belief before the request from MCA1. (B) T_tot is large when MCA2 formulates the self-belief after the request from MCA.

5.5 Self-Belief Formulation Time

The self-belief formulation time T_bf is the sum of the time taken for the accumulation of the behavior parameters to produce the behavior (T_bh), the observation formulation time (T_ob), and the belief generation time (T_BL):

5.6 Belief Exchange Time

The belief exchange time T_bx is the sum of the time taken by the REQ and REP model to request for the belief and response T_req and T_rep, respectively:

5.7 Average Belief Formulation Time

The average belief formulation time T_avg gives the average time taken for belief formulation over a node in unit time:

Tavg = ∑i = 0n(λbhi × Ttot),

where λ_bhi represents the capturing rate of the behavior parameters bh_i in a session for all values of i.

5.8 Behavior Capturing Time

The behavior capturing time T_bc is the sum of the time involved in capturing the behavior parameters from the external world and from session history:

where bh_i is the captured behavior parameter. Thus, the behavior generation time T_bh is given by

where m_j is the number of times a behavior parameter bh_i is captured.

The observation generation time can be represented as

Finally, the belief generation time can be represented as

5.9 Convergence Time

Convergence time T_con is the time interval between the request and the location of availability of required resources and is given by

where T_rreq is the resource availability request propagation time, T_tra is the time taken in tracing the resources through the nodes record, and T_tot is the time taken in belief formulation on resource availability.

According to Equation (12), MCA takes O(n²) time for self-belief formulation as Equations (16)–(18) takes O(n²) time. The total belief formulation can be done with O(n²) complexity according to Equation (11). The average belief formulation time will have O(n) complexity.

6.1 Test Bed Description

The evaluation of CAOSR is done on a test bed of a post-disaster scenario. The pooling of resources is done in a completely or partially collapsed communication infrastructure formed amid the affected area, as shown in Figure 9. The environment comprises 200 wireless mobile nodes embedded with the MCAs designed using agent factory, based on 802.11b configured in ad hoc mode, each with a data rate of 2 Mbps and random node degree. The nodes’ mobility is configured to a random waypoint with a communication range of 60 m and pause time ranging from 0 to 150 ms. The nodes are running with the Linux kernel on the IP networking protocol stack that are essential in studying the scenario, of which about two-thirds are mobile nodes such as a rescue vehicle (V₁, V₂, …, V_n) equipped with GPS-enabled IP-based netbook machines with IEEE 1609 connection and with CBR traffic. The nodes belonging to the rescue team (R₁, R₂, …, R_n) and bystanders (S₁, S₂, …, S_n) have a reduced communication range of 10–30 m and are equipped with notepads, smartphones, and tablets. The remaining nodes are static for all transmission range such as survived infrastructure (I₁, I₂, …, I_n), vehicle controllers (C₁, C₂, …, C_n), and navigation controllers (A₁, A₂, …, A_n) emulated through Mac laptops with wireless cards. The registered devices will attach to the node’s resource pool and be mapped with its class of resource and will then update the RR. The test bed gives various possibilities of a node coming in opportunistic contact (nodes in the same direction, toward same locations, executing the same task) with other registered nodes for resource pooling. It uses the RFC 2501 [25] mobile IP networking wireless protocol for node interaction. For instance, V₅ coming in opportunistic contact with R₂ can share the communication channel required for belief exchange.

Figure 9:

Test Bed Setup for Evaluation of Post-Disaster Scenario.

Initially, only the surviving infrastructure (static nodes) are positioned around the affected location. With time, the number of nodes increases due to the availability of the disaster management teams and bystanders helping the triage. The plot given in Figure 10 shows the increase in the number of nodes with time. With the increase in node density around the affected location, a resourceful contact opportunity increases.

Figure 10:

Number of Nodes vs. Time (in ms).

The plot in Figure 11 shows both homogeneous and heterogeneous opportunistic contacts. It can be observed that the number of heterogeneous opportunistic contacts are more than the number of homogeneous contacts. With the probability of having more heterogeneous opportunistic contacts than homogeneous, a node gains the benefit of using more resources than it has. Thus, a better reliability can be seen in the resource availability. The plot in Figure 12 shows the dependency of reliability over opportunistic contacts. The increase in the percentage of resourceful opportunistic contact will improve the reliability of the scheme.

Figure 11:

Number of Contacts vs. Number of Nodes.

Figure 12:

Reliability vs. Total Number of Node Contacts.

The plot given in Figure 13 shows the influence of opportunistic contacts on resource availability. The nodes on opportunistic contact can coordinately determine the availability of resources, which forms a medium for dispersion of information throughout the affected area. An agent program running on an MCA captures the required variables to determine the resource availability. Therefore, with an increase in opportunistic contact, the percentage of resource availability also increases.

Figure 13:

Resource Availability vs. Opportunistic Contact.

The plot in Figure 14 describes the effect of node density on the generation of four classes of beliefs. With an increase in node density, resource availability increases and the agent will be able to generate beliefs with reduced belief formulation time. The patron class of beliefs dominates, and the slack class of beliefs degrades with time. This result reflects that more resource can be pooled up with opportunistic contact to manage the critical scenarios dynamically and with reliability.

Figure 14:

Belief vs. Node Density.

The plot given in Figure 15 shows the effect of pause time on the resource availability convergence rate. It can be observed that a shorter pause time of the scenario, i.e. higher node mobility, leads to increased resource availability convergence rate. The higher mobility rate of the nodes provides more opportunistic contacts, which helps in faster resource location and hence higher convergence rate. Thus, at higher mobility, the network becomes more dynamic.

Figure 15:

Node Density vs. Pause Time.

Figure 16 shows the plot of the percentage of belief formulation failure rate with time. As the resource availability increases, the process of belief formulation is to be increased at the same rate. However, the resource availability convergence rate is also dependent on the mobility of the nodes within the application scenario. Initially, the mobility will be less due to the collapsed pre-existing infrastructure. Therefore, the number of opportunistic contacts will be fewer. Thus, the failure rate is high at the beginning, and with time, it will decrease due to the increase in mobility.

Figure 16:

Failure Rate (in %) vs. Time Interval.