Sage Journals: Discover world-class research

Abstract

The redundancy of the data is an active research topic. While an agent works in a multi-agent system, the number of messages between them increases. This is due to the fact that the functionalities data depends on other agents in terms of functional requirements. Typically, only one agent in a multi-agent system is responsible for accessing a database instead of replicating the database on each agent. A database is stored on multiple agents rather than a single agent to avoid a single point of failure. In this approach, the system has a higher load because one agent is responsible for all agent queries and must send duplicate messages to multiple agents, resulting in redundant data. In this research, we present Multi-Agent System for Commodity Data (MASCD) framework, the multi-agent system based communication using the distributed hash system, to reduce data redundancy in multi-agent system communication. Our anticipated method demonstrated how we divided the database names and efficiently distributed data to each agent. The database splitting is based on manufacturer names or product names. We utilize a table based on prime numbers. Through the hash function, we ascertain the index of the agent granted access to the relevant data. Each agent is accountable for its data. We use a Distributed Hash Table for efficient querying that stores data as key-value pairs. Each agent maintains a Finger Table containing the next and previous nodes for agent communication purposes. Using FIPA messages, we demonstrated how an agent could interact optimally. In conclusion, we illustrate the application of the proposed approach through a case study of mobile phones and university information systems.

Keywords

Multi-agent systems distributed hash table agent communication FIPA communication

ï»¿

1. Introduction

Today, data redundancy is a significant research topic in multi-agent systems. This is because the quantity of messages sent between agents in a multi-agent system rises dynamically with time. In addition, the system’s workload is increased because a single agent must respond to all agent queries. A network of autonomous software agents working together to accomplish tasks and communicate with one another in a multi-agent system. Agents make decisions independently. Agents must work together and coordinate their efforts because they have unique goals and strategies. Agents take action based on various factors, including what they observe in their immediate surroundings and how they interact with other agents. The areas of robotics, autonomous driving, and commercial games have all shown significant interest in multi-agent systems. In [1], as the size of the problem increases, the complexity of the modeling and calculation tasks increases significantly. As a result, we cannot deal with these scenarios by employing a centralized method. Because of this, we made use of a multi-agent system in order to deal with the various scenarios. One of its most significant advantages is that the multi-agent system can function in a decentralized setting [2]. Inter-agent negotiations are conducted through Agent Communication Languages (ACL). In [3] there are two of the most familiar agent communication languages, i.e., KQML and FIPA. It is based on speech-act theory. KQML is a usual agent communication language where agents can freely share and discuss information and ideas. Messages in KQML follow a specific protocol. It’s a protocol for dealing with messages that uses the principles of speech acts theory [4].

Therefore, communication significantly impacts the multi-agent system, but an issue that arises in multi-agent system is the overhead associated with the communication. In a multi-agent system, data redundancy is the problem’s primary source, resulting in increased storage needs. For instance, a self-driving automobile is currently making its way up a mountain, and the front camera of this car can take pictures of the trees on both sides, which results in the interchange of noisy and comparable data throughout the observation, leading to data redundancy. In [5] the technique of temporal massage decreases data redundancy in agent communication. For this model, buffers are used. The first is the sender buffer, while the second is the receiver buffer. Both the sender and the receiver may utilize buffers to hold messages temporarily. The sender only sends a new message when significant new information is compared to the previous message. This model likewise includes a message encoder and a local action generator. However, the temporal complexity of finding the message increases by a factor of O(n) in the event of transmission loss [5].

Our primary research objective is to decrease the time to resolve queries and eliminate data redundancy in real-time. Data redundancy proves to be a significant hurdle in agent communication, preventing the multi-agent system from functioning optimally. Agent database storage strategies are expanding as a result of redundancy. The overall performance of multi-agent system is dynamically affected by the growing volume of messages sent and received between agents over time. In prior methods, access to a database was typically delegated to one of the agents working in a multi-agent system. Because one agent is responsible for every other agent’s query and must transmit the same messages to multiple agents, which causes data duplication, a centralized approach leads the system to have a higher burden than a decentralized approach. This research provides Multi-Agent System for Commodity Data (MASCD), an approach to decreasing duplicate data in multi-agent system communication. To implement Distributed Hash Table (DHT), we first partitioned the database into smaller subsets and assigned each agent to handle a smaller portion of the database. This method is employed when there is a lot of data since it makes query-finding data easier. No prior research has recommended a method that utilizes a hash-table database for this specific issue. The main issue is finding a way to reduce the volume of messages without impairing the multi-agent system’s performance as a whole.

The remaining paper structure is defined in the following section. The first section provides an introduction and background information. In Section 2, a related work is presented. A distributed hash table is the main component of the MASCD framework which is explained in Section 3. In Section 4, case studies are provided to exemplify our suggested approach. These real-world case studies illustrate the viability of our suggestion and are based on actual basic cases. This chapter concludes the thesis and offers suggestions for future research.

2. Literature review

In references [6, 7], KQML and FIPA are depicted as prevalent languages that facilitate agent communication and collaboration. They endorse the effective dissemination of knowledge and ensure a seamless exchange of information. Messages in KQML are based on a protocol. That’s because it’s a method for transmitting and receiving messages. Agents can share information via a communication facilitator. They proposed new features for e-commerce agents to carry out. Their research’s primary objective is to lessen the network overhead burden. In [8], an impactful multi-agent communication approach is provided because of the significant problem in multi-agent reinforcement learning: limited bandwidth; therefore, agents should generate and send only informative messages. In [9], the primary consideration in a multi-agent system is how effectively information is communicated. Agent Communication Languages (ACL) significantly impact multi-agent systems, but they incur overhead. Two-tiered communication architecture was proposed for efficiency purposes. In [10], multi-agent system describes how smart sensor nodes in a wireless sensor network should act. When collecting data, it is essential that sensor nodes are close to each other and that there is a high chance of redundancy. Mobile agents are used to gather data for this reason. In order to collect data and increase energy efficiency, [11] proposed the Itinerary Energy Minimum Algorithm (IEMA). We can ascertain the number of mobile agents by presenting a genetic framework that accounts for the time complexity. Multiaccess edge computing (MEC) is a technology that enables self-driving vehicles to compete in time-sensitive and data-intensive computational activities. In this method, automobiles produce an excessive amount of redundant data. Within the framework of the proposed technique, the minimum possible number of pictures is transferred to MEC servers [12]. In [13], they presented a new strategy that they referred to as message dropout. To improve the performance of multi-agent deep reinforcement learning in two different scenarios. In [14], they investigated a few value-based multi-agent strategies. We look into value-based solutions for multi-agent reinforcement learning (MARL) problems in the recently popularized centralized training with decentralized execution (CTDE) regime.

As stated in [15], many agents can learn to work together towards MARL by using mutual knowledge between actions to regularise the cumulative return. For a given Maximum Mutual Information (MMI) regularized objective function, we propose a manageable bottom-bound strategy. To solve the problems that have been occurring with traffic light control systems, a multi-agent communication and action rectification (MaCAR) Framework has been presented in [16]. This strategy maximizes traffic efficiency while making the most of the available road space and helps to prevent road rage incidents. This framework facilitates the active communication of agents with one another. In multi-agent systems, communication between agents is an essential component. They suggested a new strategy in the context of a multi-agent system founded on Khalimsky’s idea. This method typically performs data routing in wireless sensor networks [17]. In [18], they explored how the formulation of control policies in a multi-agent system is affected by the choices made about communication. The authors proposed an Markov decision process (MDP) based distributed framework to model the expenses of sending messages. In [19], to address this problem, they designed a dynamic event-triggered communication mechanism (DECM) for coordinating conversations across agents. It is crucial to cut down on extraneous information and non-transferable between different agents for efficiency. In [20], a novel platform based on the multi-agent system is presented. They employed a Jabber protocol to enable instantaneous communication between applications (XMPP – Extensible Messaging and Presence Protocol). This system complies with FIPA and provides a straightforward technique for creating agents employing this modern communication method. In [21], with the use of a multi-agent system application, they show how to create a smart grid efficiently. The appliances in a smart house are modeled after autonomous agents. The HEMS (Home Energy Management System) was created with the help of multi-agent system. It opens the way for networked homes to share information and collaborate. In [22, 23] an alternative cognitive radio spectrum sensing framework based on a multi-agent architecture has been proposed for 5G wireless communication systems. The waste of materials and excess energy use are both mitigated by this system. Furthermore, it prevents agents from communicating the same information twice.

In [24], they employed consensus strategies to have agents work together on unmanned aerial vehicles. Their framework enables agents to handle various task assignments in a dynamic environment with less communication overhead and a new storage mechanism. In [25, 26], they combine the two methods of control in order to study the bounded consensus issue in multi-agent systems. It uses quantized control and event-triggered control, which rely on an undirected graph, to lighten the communication load and get around network limits like bandwidth limitations. Event-triggered quantized communication protocols are developed to achieve bounded consensus in multi-agent systems. In [27], they presented an event-triggered distributed subgradient method for tackling convex optimization problems. These issues are modelled as the union of the convex cost functions of all involved agents.

In [28], that study provided a load-sharing control finding method for an inverter-based microgrid. It adopts a decentralized multi-agent system strategy that emphasizes effective communication. After an event, the agent will notify any other agents in the area of its current status. Data exchange between agents is impossible if this condition is not met. In [29], for systems with multiple agents performing random switches, they suggest an adaptive optimum decision solution. Random switching is widely used in the realm of communication. It offers a precise online decision-making solution for several agents in a short amount of time. It’s a smart strategy for minimizing time complexity. In [30], for combining FIPA performatives with Timed-Arc Petri-Nets (TAPN) in RESTful, event-driven, or event-driven distributed multi-agent systems, a formal paradigm is proposed – real-time multi-agent systems (RTMAS). A multi-agent system with a time limit was the topic of their conversation. While TAPN has been used to describe FIPA in prior works, the lack of timing is the primary problem in RTMAS. According to [31], there are two different categories that multi-agent navigation can fall into. Both centralized and decentralized methods are included in this category. To counter multi-agent navigation, many types of sensor noise are utilized. They created a framework to create an efficient distributed multi-agent navigation collision avoidance policy.

In [32], The cost of communication is decreased through numerous algorithms. Our primary concern is guaranteeing the network’s long-term survival. Improved estimation accuracy and reduced communication costs are achieved by employing a distributed version of the neighbor-partial diffusion least-mean-square algorithm (NPDLMS). Multi-agent robotics needs communicative drones. Centralized and decentralized mission control were used. We employ centralized architecture for real-time monitoring. Otherwise, a decentralized application [33]. In [34], they introduced a communication protocol called Communication Collective to decrease network latency. Due to rising urban populations and reducing automobile costs, traffic congestion is a persistent issue in many countries [35]. The signal control system helps choose a “green time” for faster network-wide transit based on limited environmental data. A geometric fuzzy multi-agent system (GFMAS) should be used for improved accuracy and performance. Utilizing cutting-edge technology, they are creating sensing applications [36]. They utilized the Gaussian process regression (GPR) method, which enables agents to anticipate the behavior of ecological systems.

According to [37], today’s society has its most significant difficulty managing large amounts of data. We employ wireless sensor networks to create data propagation networks that include many sensors. One of the main functions of sensors and big data is to keep an eye on a process, gather data, and send it to a central location for processing. In [38], the author presents A focused communication architecture for multi-agent reinforcement learning. In this concept, agents interact in partially viewable settings to learn which messages to send and to whom. In [39], two distinct contributions can be found in their work. First, they show that policy gradient iterations can be used outside of Markovian frameworks. Second, break down the problem into elements of a policy for wants.

Figure 1.

Multi-Agent System for Commodity Data (MASCD) framework.

3. Proposed Multi-Agent System for Commodity Data (MASCD) framework

This section outlines the Multi-Agent System for Commodity Data (MASCD) system we developed to reduce data redundancy in multi-agent system communication. Our entire framework is depicted in Fig. 1. There are three multi-agent systems in our framework. Each multi-agent system has a different domain model. We handle a broad spectrum of product data in our Multi-Agent System for Commodity Data (MASCD). Based on the specific needs of a query, our agents are capable of interacting efficiently with other multi-agent systems and communicating effectively amongst themselves. This applies whether a query originates within our system or externally from another multi-agent system. Each multi-agent system consists of a range of modules, which we delve into extensively in the following section. It is explicitly stated that agents must utilize FIPA messages for communication. In the MASCD framework, we use a DHT to reduce data redundancy in agent communications. To accomplish this, we have segmented the database into smaller subgroups and deftly disseminated the data of these subsets across all agents. Our approach accommodates two scenarios in which a database is divided. In scenario 1, we divide the database according to the product names, and in scenario 2, we do the same according to the manufacturer names. The hash function is in charge of data distribution. We work with a variety of commodities or product data in MASCD. Agents can proficiently interact with other multi-agent systems depending on the query’s specifications, whether it originates from within or outside the multi-agent system.

The division of the database into chunks is depicted in Fig. 2. Data is given to the hash function, which converts it into keys, each intended for a different agent. The key-value pairs are kept in the DHT. As a result, each agent is responsible for a particular key and collection of information. Agents receive the data according to the product name. The agent database contains a large number of entities. However, we prioritize the main entity and distribute information to various agents based on the product name. For instance, we prioritize the student as the main element in the student database. In order to do this, a DHT is used. A decentralized method of key-value pair storage for data is the DHT. The DHT makes it easier to extract data from enormous volumes of data. A DHT node is analogous to an agent accountable for the keys and values with which it is associated. With this decentralized approach, agents create a system devoid of centralized power. Each agent keeps track of the addresses of the subsequent and preceding nodes in a finger table.

Figure 2.

Distributed hash table (DHT) representation.

The addresses of the successor and predecessor nodes are stored in a finger table on each node of a distributed hash table. The agent determines whether data is present before responding to a request for data. In order to find out if the requested id and the subsequent or preceding id are the same when an agent is unaware of data, it checks the finger table. If the finger table does not contain a match, we count the number of instructions in both the clockwise and backward directions. The query is transmitted to the subsequent node if the clockwise steps are less than the number of anti-clockwise steps; otherwise, the question goes oppositely. Since agents are connected circularly, the finger table of an agent’s successor and predecessor nodes is updated whenever an agent enters or exits the system. Assume that each agent has access to a database subset, which reduces data redundancy and storage space.

$\displaystyle\textit{successor}=\sum_{i=1}^{N}\sum_{j=i}^{j<i+1}j=j+1\bigcup% \textit{predeceesor}=\sum_{i=1}^{N}\sum_{j=i}^{j>j-1}j=j-1$ (1)

Equation (1) illustrates how each node’s Finger Table contains the address of the subsequent ID and the ID of its preceding nodes – the external loop cycles through 1 to N, where N represents the total count of agents. Simultaneously, the internal loop manages ‘j’ as the current agent and records the addresses of the following and preceding agents in the current agent’s Finger Table. The subsequent section describes the various elements that constitute the proposed DHT framework.

3.1 Optimization approach

In our paper we have used the word optimized in terms of an efficient approach. The proposed approach ensures that rather than overloading a single agent with queries including repeated queries we use a distributed has table based approach to distribute the database. We can formally present the message redundancy problem in multi-agent systems as.

Let D be the set of all data items in the multi-agents system.

Let A be the set of all agents in the system.

Let $X_{ad}$ a binary variable indicating whether agent a has a total copy of data item d.

Let $C_{ad}$ be a binary variable indicating whether agent a communicates data item d to other agents.

The goal is to minimize the overall data redundancy subject to certain constraints.

3.1.1 Objective function

$\displaystyle\textit{Minimize}\sum_{a\in A}\sum_{d\in D}C_{ad}$ (2)

Equation (2) represents the sum of communication indicators across all agents and data items. Minimizing this sum encourages reducing redundancy by minimizing the number of times data items are communicated.

3.1.2 Communication overhead constraints

$\displaystyle C_{ad}\leqslant X_{ad},\forall a\in A,d\in D$ (3)

Equation (3) ensures that an agent can only communicate a data item if it has a local copy of that item.

3.1.3 Data availability constraints

$\displaystyle\sum_{a\in A}X_{ad}\geqslant 1\forall d\in D$ (4)

Equation (4) ensures that each data item is available at least once in the system.

3.1.4 Communication overhead constraints

$\displaystyle X_{ad},C_{ad}\in\{0,1\},\forall a\in A,d\in D$ (5)

Equation (5) imposes constraints on the total communication overhead for each agent, where B is a predefined threshold. This formulation is a simplified representation and can be adapted based on the specific characteristics and requirements of the multi-agent system. The binary variables $X_{ad}$ and $C_{ad}$ are crucial in capturing the presence of data on agents and the communication of data between agents, respectively. The objective function and constraints aim to balance the need for data availability with the goal of minimizing redundancy in data storage and communication.

3.2 Modules in our framework

Our framework consists of various modules, each of which is covered below.

3.2.1 Directory facilitator (DF)

The focal point of our framework is the DF. It offers services related to the yellow pages, suggesting that DF gives the agents access to a service repository. Service descriptions are created by agents and stored in the DF. Users can obtain reports on the requested service by accessing the DF. The DF provides services that offer platform agents a list of skills and abilities. The directory facilitator’s Knowledge Base is a digital repository for service descriptions. DFAgentDescription objects contain the details of the DFAgent’s description. In our design, agents’ primary objective is to interact with one another as effectively as feasible. Each agent is accountable for their data since we wish to eliminate data redundancy; as a result, agents must first specify their services in the directory facilitator. Each agent’s essential duties include effective communication, keeping track of particular key and detailed information, and maintaining the finger table if a node joins, leaves, or fails. The Knowledge Base is in charge of storing the agent’s services. DF also has a search feature for getting data from a specific agent and registers and deregisters for storage.

3.2.2 Agent management system (AMS)

Lifecycle management is one of the services that may be obtained using the Agent Management System (AMS) [44]. The creation, deletion, and restart processes are included in the management of the life cycle of the agents. In addition, it is in charge of delivering services to agents, such as name lookup and registration, and it is responsible for authenticating agents. The “white pages” function of the platform is provided by the AMS, which is responsible for controlling the lifecycle of the management system used by the agents. It offers the “white pages” service, which monitors the current locations of mobile agents and sends that information to users. Therefore, the AMS is utilized to create an agent for some purpose or delete some agents in the event that we so desire.

3.2.3 Agents

A software or computer program that makes decisions on behalf of its users is known as an agent [43]. It is autonomous or partially autonomous. Each agent has specific plans and objectives, and successful interaction requires collaboration and coordination. In our proposed design, we seek to facilitate effective agent communication. Intelligent agents possess reactiveness, proactivity, and sociality; hence, the agent must be intelligent. They can make quick decisions based on nearby observations and interactions with other agents.

3.2.4 Distributed hash table (DHT)

The core component of our proposed architecture is a DHT. It searches enormous amounts of data in a decentralized storage system to locate information. By using this technique, data duplication between agents is minimized. Data redundancy is the main problem in agent communication, which contributes to system inefficiency and storage increase. We use a distributed hash table as a result. The database is partitioned into subgroups in our suggested manner, and the data from each subset is then given to various agents. When a request for a date is made, it is sent to a random agent; if that agent cannot fulfill the request, it is sent to the following or prior agent. Each node is in charge of this data. In the context of DHT, a chord is a specific type where each agent maintains the node addresses before and after it. We assume that data is kept in some repository. We talk about minimizing redundant information in agent communication. Our proposed approach offers a significant advantage of data partitioning, wherein each agent handles a specific data set, eliminating data redundancy within the system.

3.2.5 Finger Table (FT)

Every node in a DHT keeps track of the addresses of the next and previous nodes through a structure known as a Finger Table. This structure makes DHTs exceptionally effective for quick searches across vast data. Access to the nodes in the Finger Table is straightforward, facilitating efficient execution of queries. Being the core element in the Chord DHT, the Finger Table plays a pivotal role in the overall functionality of the Chord.

3.2.6 Agent communication channel (ACC)

The ACC facilitates two-way communication between agents operating on different systems. Agents in the same location can communicate with one another via notifications. An agent communication channel allows agents to exchange information with one another. Achieving objectives and increasing the system’s effectiveness calls for cooperation between agents.

3.2.7 Agent communication language (ACL)

Negotiation between agents takes place through the use of ACL. The most well-known ACLs are KQML and FIPA, based on speech-act theory. These languages are deployed on JADE and AnyLogic, respectively.

3.3 Query in a distributed hash table (DHT)

In our framework, we postulate that each agent has a section of the database at their disposal, communicating through a messaging transit system. These database subsets are labeled D1, D2, and up to Dn. Every agent in our setup holds a finger table, providing addresses of both its predecessor and successor. When a request reaches an agent without an acknowledgment symbol, steps are calculated in clockwise and anti-clockwise directions. We present below several reasons for choosing a particular course of action. There are two scenarios for counting steps in the clockwise direction: Case 1 and Case 2 described in Fig. 3. We decide between these two cases depending on the conditions for counting clockwise steps. Meanwhile, we consider Case 3 and Case 4 for counting steps in the anti-clockwise direction.

Figure 3.

Distributed hash table (DHT) queries.

In cases no 3 and 4, choosing the route involves going the other way around. The source ID in the scenarios mentioned above refers to the identifier associated with the origin of the data request. The requested id is the hash id of the ordered product data or a destination id queried by some user. The number of steps in a multi-agent system determines the maximum no of the agents. We must decide if the requested id is larger than or less than the source id in order to count the number of clockwise steps between them. There are unique formulas for each event, and depending on the circumstances, we choose formulas and count the number of clockwise and anti-clockwise steps. If the count of clockwise steps is lower than the count of anti-clockwise steps, the inquiry progresses forward. Otherwise, the inquiry goes in a backward direction. We are currently looking at a mathematical illustration of data retrieval from the agent. Let A represent the group of agents A $=$ Ag1, Ag2, Ag3 and M is the collection of messages exchanged by agents M $=$ m1, m2, and m3.

$\displaystyle k_{r}\rightarrow\{K_{i},K_{i+1}\}\cup\{K_{j},K_{j-1}\}$ (6)

Equation (6) shows that an agent retrieves data from the Finger Table when he cannot satisfy a request for data. $K_{i}$ and $K_{i+1}$ represent the agents in a forward direction, and $K_{j}$ , $K{j-1}$ represents the agents in a backward direction. $K_{r}$ is an agent who receives a request to find the data. Let’s say the Finger Table is unable to locate the data. In that situation, he counts the steps in clockwise and anti-clockwise directions to find the shortest route between the destination and the initial id. Let R stand in for the set of requests.

$\bm{R\to K_{i}}$ : (If Agent K cannot fulfill a request, it verifies whether the requested Id matches that of subsequent or predecessor node).

$\bm{R\to K_{j}}$ : (In the absence of an acknowledgment from node $k_{i}$ , a request goes backward if the count of anti-clockwise steps is less than the number of clockwise steps, which is $K_{j}$ , $K_{j-1}$ and continues until reaching the destination point.)

$\bm{R\to K{i+1}}$ : (A request will be routed clockwise (to nodes $K_{i}$ , $K{i+1}$ , and so on) until it reaches the destination point whenever the number of steps taken in one direction is greater than the number of steps taken in the other.)

The results of our suggested strategy are shown in Fig. 4. Each node has a distinct id, host, and port, as well as storage for the addresses of its predecessor and successor nodes. A finger table on each node in a distributed hash table contains the addresses of successor and predecessor nodes. When an agent is asked for data, it checks to see if it has any. An agent checks the next and previous nodes if it has no data, and if neither can answer the query, it checks a random node in either a clockwise or anti-clockwise manner.

Figure 4.

Distributed hash table (DHT) representation.

Whenever a new agent joins or leaves the system, we ensure to update the finger tables of its successor and predecessor nodes. This approach reduces data redundancy and minimizes storage requirements as each agent only possesses a portion of the database. We implement the distributed hash table using Node.Js. The code snippet in Table 1 showcases the implementation of the distributed hash table, incorporating the addresses of the subsequent and predecessor nodes.

Table 1

Implementation of distributed hash table (DHT) in node js

interface Node {

id: number;

host: string;

port: number;

}

interface FingerTableEntry {

start: number;

successor: Node;

predecessor: Node;

}

class distributed hash table

{

id: number;

host: string;

port: number;

fingerTable:

Array

<\textit{FingerTableEntry}>

=

[

{

start: null,

successor: NULL_NODE,

predecessor: NULL_NODE

}

];

constructor ({id, host, port}) {

this.id

=

id;

this.host

=

host;

this.port

=

port;

}

encapsulateSelf (): Node {

return {

id: this.id,

host: this.host,

port: this.port

};

}

async setSuccessor (

const successorCandidate

=

message.request;

(

successorCandidate.id

==

null &&

successorCandidate.host

==

null &&

successorCandidate.port

==

null

)

{

this.fingerTable[0].successor

=

successorCandidate;

}

console.log ("setSuccessor:

new successor

=

async getFingerTable (call: {

write: (arg0: {index: number;

node: Node})

\Rightarrow

void;

this.fingerTable.forEach

(fingerTableEntry

\Rightarrow

{

index: fingerTableEntry.start,

node: fingerTableEntry.successor

}); });

call.end (); }

async setPredecessor (

message: {request: Node},

callback: (arg0: any, arg1: {})

\Rightarrow

void)

{if (DEBUGGING_LOCAL) {

console.log ("setPredecessor:

Self

=

", this.encapsulateSelf ());

console.log ("setPredecessor:

Self’s original predecessor

=

this.predecessor); }

this.predecessor

=

message.request;

async function loadData {

const response

=

await fetch ("./data");

const myJson

=

await response.json ();

const nodes

=

Object.values (myJson)

.map (({id, host, port, …rest}, index)

\Rightarrow

({

id: ‘${host}:${port}’,

3.4 Efficiently data distribution

Previous approaches demonstrate that each agent contains the same database, which provides redundant data. We present the distributed hash table technique for removing redundant data. With this technique, the database is split up according to the product description, and each agent gives the data equally. This is accomplished by reviewing Tables 2 and 3, applying a hash function, and mapping the value to the appropriate index after converting the product name into a numeric number. For instance, our database includes information from several companies. Based on product names, we distribute product data to each agent equally. Using Table 2, we convert each product name to a numeric value by applying a hash function. The following algo is used to elaborate on the code that is described in Table 1.

: Description for Table 1[1]

DHT Working $a, b$ System Initialization Create Node $\leftarrow$ id, host, port Create FingerTable $\leftarrow$ adresses of successor, predecessor Initialize DHT $\leftarrow$ connected agents in circular way Create FingerTable $\leftarrow$ adresses of successor, predecessor set successor $\leftarrow$ keep the address of next node set predecessor $\leftarrow$ keep the address of previous node query for finding data each agent $a$ in DHT $\textit{condition}=\text{True}$ return data Break continue

The following expression represents our hash function. C is the hash table’s size, while y is an integer number. The agent’s index, which contains the product data, is obtained once the modulus has been calculated.

$\displaystyle h(y)=y\%C$

We create a second hash function that makes use of the first hash function and the parameter c in the event of a collision.

$\displaystyle h^{\prime}(y)=(h(y)+x)\%C$

C is the size of the hash table, and x is the incrementing value used to create an empty index. We’ll talk about the selection criteria for x right now. It is best to choose x to be 1. However, if we want to locate an empty slot effectively, x and C must be reasonably prime for resolving collision. We decide on c as the winner. Applying the hash function yields an index of the agent in charge of a given data set. Using Tables 2 and 3, we translate strings into numerical values.

Table 2
a-z prime alphabets

1009 $<=$ a	1013 $<=$ b	1019 $<=$ c	1021 $<=$ d	1031 $<=$ e	1033 $<=$ f
1039 $<=$ g	1049 $<=$ h	1051 $<=$ i	1061 $<=$ j	1063 $<=$ k	1069 $<=$ l
1087 $<=$ m	1091 $<=$ n	1093 $<=$ o	1097 $<=$ p	1103 $<=$ q	1109 $<=$ r
1117 $<=$ s	1123 $<=$ t	1129 $<=$ u	1151 $<=$ v	1153 $<=$ w	1163 $<=$ x
1175 $<=$ y	1181 $<=$ z	space- $>$ 1187

Table 3

A-Z & 0–9 prime alphabets

823 $<=$ A	827 $<=$ B	829 $<=$ C	839 $<=$ D	853 $<=$ E	857 $<=$ F
859 $<=$ G	863 $<=$ H	877 $<=$ I	881 $<=$ J	883 $<=$ K	887 $<=$ L
907 $<=$ M	911 $<=$ N	919 $<=$ O	929 $<=$ P	937 $<=$ Q	941 $<=$ R
947 $<=$ S	953 $<=$ T	967 $<=$ U	971 $<=$ V	977 $<=$ W	983 $<=$ X
733 $<=$ Y	739 $<=$ Z	701 $<=$ 0	709 $<=$ 1	719 $<=$ 2	727 $<=$ 3
733 $<=$ 3	739 $<=$ 4	743 $<=$ 5	751 $<=$ 6	757 $<=$ 7	761 $<=$ 8

As proposed by this paper, the tables mentioned above assign each letter from ‘a’ to ‘z’ a unique prime value reminiscent of ASCII encoding. As a result, string values are transformed into numeric values using this Table [41]. A distributed hash table is then used. We remove redundancy and equally distribute data to each agent using our suggested strategy. The DHT methodology is shown in Fig. 4. Each agent is liable for a particular key and value and is in charge whenever a query occurs. Each agent retains their data in this technique, which eliminates data redundancy.

4. Demonstration of the proposed Multi-Agent System for Commodity Data (MASCD) architecture

In our framework, a DHT is utilized to store the addresses of the nodes that precede and follow any given node. We segment the database according to the names of the products, and each agent is in charge of a particular set of information. The corresponding agent is responsible for responding to requests for specific data. A random node is queried by the user for data; if the node has data, it should respond; otherwise, the query proceeds in either a clockwise or anti-clockwise manner by using the formulas we mention in the query section. The forthcoming case examples illustrate our proposed approach, which we delve into more deeply in the subsequent section.

4.1 Case study of e-commerce agents using mobile phones

Next, we’ll discuss the practical application of our framework, providing an example to demonstrate its immediate effectiveness in reducing redundancy. We use Infinix, Tecno, Nokia, and other manufacturing companies as examples. We have a database that stores the product information for each company. In this case study, examples of mobile phone products are provided. In [2], each agent shared the same database, resulting in redundant data. We present the DHT technique for removing redundant data. Let’s say three companies’ mobile phone information is in our database. Examples of firms that store data in our database include Tecno, Infinix, and Nokia. We start by talking about Infinix Corporation data. Infinix introduced various products, including the Note 10, the Note 10 Pro, and the Note 11. Based on the names of mobile phones, we distribute data. The hash function was explicitly developed to address this need. The function below depicts this hash function.

$\displaystyle h(y)=y\%C$

The hash table size is represented by $C$ , while $y$ denotes an integer. After applying the modulus operation, we derive the index of the agent that retains the manufacturer’s product data. This hash function is utilized to compute the hash index for each product. We distribute data based on string-based mobile names or descriptions. For instance, the flagship product of Infinix is Note 10. Initially, we convert Note 10 into its numerical equivalent using Table 2, after which the hash function is applied.

Assume that the hash table is 14 bytes in size. A prime value from Table 2 has been assigned to each letter. The value of the letter N is 911, the value of the prime number “o” is 1093, the value of the letter “t” is 1123, and the values of the letters “e,” “1,” and “0” are 1031, 709, and 701, respectively. Where the space has a value of 1187 respectively. The sum of all these alphabetic numeric values is now calculated as y.

$\displaystyle y=911+1093+1123+1031+1187+709+701\Rightarrow 6755$

4.1.1 Calculating hash indexes

Having applied the hash function and incorporated the value of ‘x’ into our equation, we are now equipped to identify the index of the agent responsible for a specific data batch.

$\displaystyle h(y)=6755\%14\Rightarrow 7$

Therefore, after modulus, we obtain index seven, which indicates that agent 7 is in charge of mobile data in great detail. Agent 7 is in charge of doing a query for this kind of data. The Note 10 Pro and Note 11 now follow the same procedure. The string is converted to a numeric number, and then the hash function is used. We calculate $y$ by adding the Note 10 Pro’s numerical values.

$\displaystyle y=911+1093+1123+1031+1187+709+701+1187+929+1109+1093\Rightarrow 1% 1073$

Now, we calculate $y$ , which is 11073. Now enter the $y$ and $C$ numbers into the equation for the hash function.

$\displaystyle h(y)=11073\%14\Rightarrow 13$

The hash function computation yielded the number thirteen, which suggests that Agent 13 is in charge of the Note 10 Pro. A specialized agent is in charge of responding to user inquiries for information regarding the Note 10 Pro. We now determine the hash index for Note 11.

$\displaystyle y=911+1093+1123+1031+1187+709+709\Rightarrow 6747$

We get the index of the agent responsible for this data after plugging the values of $y$ and $C$ into the equation for the hash function.

$\displaystyle h(y)=6763\%14\Rightarrow 1$

Agent 1 is responsible for responding to queries about retrieving Note 11 data whenever they occur. Therefore, there are three agents responsible for Infinix company data. Tecno introduced mobile devices such as Spark 9 and Spark 9 Pro. Referring to Table 2, we first convert each product to its numeric value. The topic of Tecno smartphones comes first. Below is a list of the Spark 9 numerical values.

$\displaystyle y=947+1097+1009+1109+1063+1187+761\Rightarrow 7173$

By entering the values of x and s into the equation above, we can get the hash function. This provides us with the agent’s index.

$\displaystyle h(y)=7173\%14\Rightarrow 5$

Agent 5 is responsible for responding to queries about retrieving Note 11 data whenever they occur. Now the Spark 9 Pro follows the same procedure. Below is a list of the Spark 9 Pro numerical values.

$\displaystyle y=947+1097+1009+1109+1063+1187+761+1187+929+1109+1093\Rightarrow 1% 1491$

Now, by entering the values of $y$ and $C$ into the equation above, we can get the hash function. This provides us with the agent’s index.

$\displaystyle h(y)=11491\%14\Rightarrow 11$

Agent eleven is, therefore in command of this cell phone. Agent eleven’s databases should contain Spark 9 Pro because this agent offers specific information. Now the same calculations are applied to Nokia 7 and Nokia 7 Plus.

$\displaystyle y=911+1093+1063+1051+1009+1187+751+1187+929+1069+1129+1117$ $\displaystyle\Rightarrow 12495$

After applying, we get an index of 7. Because this index is not empty, there is now a collision. We added another hash function to fix the above collision.

$\displaystyle h^{\prime}(y)=(h(y)+x)\%C$

The function we calculated before to ascertain the hash index of a particular agent is called $h(y)$ . The increment size is $x$ . $x$ and $C$ must be roughly prime in order to resolve collision. Prime numbers that have just one element in common, or a relative prime, are 1. As a result, this is the best technique for figuring out collision. Our hash function’s increment size, $x$ , is set to 1, so we set it to 1. Put the $x$ and $C$ values into our suggested hash function.

$\displaystyle h^{\prime}(y)=(7+1)\%14\Rightarrow 8$

As a result, we discover the agent’s index eight is responsible for Nokia 7 Plus. Agent 8 is in charge of answering inquiries about the Nokia 7 Plus smartphone.

Let’s say a user requests information on Note 11 from agent 11 in Fig. 5. First, we check the phone’s hash value and then search the finger table to see if it has the requested data. Using a formula, agent 11 counts clockwise and anti-clockwise steps to reach the final destination. We find the number of steps by implementing the formulas described in the query section until we reach the target; the query goes to the successor node if clockwise steps are less than anti-clockwise steps. Until we locate our destination node, the query moves to predecessor nodes if clockwise steps exceed anti-clockwise steps. In our search, we look for successor nodes, 12, 13, 0 and 1 agents, and Note 11. For Note 11 data, we discover 1, a hash value of the requested query, using a finger table. Figure 5’s second query is for Note 7 data from Agent 5. Note 7’s hash value is 9, our target. Applying the same process as before, the query is forwarded to succeeding nodes until it reaches its final destination. The query is sent to 6, 7, 8 and 9 in our example (our destination). Node 9 is responsible for particular data.

Table 4
Split database on the base of product name

Agent 1	Note 11
Agent 5	Spark 9
Agent 7	Note 10
Agent 8	Nokia 7 plus
Agent 9	Nokia 7
Agent 11	Spark 9 pro
Agent 13	Note 10 pro

Figure 5.

Query mechanism for case study.

When a query is executed, a particular agent is responsible for a specific data set, as the table above indicates.

4.1.2 FIPA messages

1) request

: (A 11) sender

: (A 12) receiver

: MASCD-FIPA ontology

: CL language

: fipa-request protocol

: ((Data: Request for Note 11 related data, or a user query.)) content

2) failure

: (A 12) sender

: (A 11) receiver

: MASCD-FIPA ontology

: CL language

: fipa-request protocol

: ((Data: query for particular data is failed. query forwarded to predecessor node)) content

3) request

: (A 12) sender

: (A 13) receiver

: MASCD-FIPA ontology

: CL language

: fipa-request protocol

: ((Data: Request for Note 11 related data, or a user query.)) content

4) inform

: (A 1) sender

: (A 0) receiver

: MASCD-FIPA ontology

: CL language

: fipa-inform protocol

: ((Data: Response with required data.)) content

The number of messages represents that, in four requests, we were able to retrieve data for Note 11. There are a total of four steps in which we responded to the query with specific data. In the initial request, Agent 11 request specific data for Note 11 from Agent 12. In the event of a failure, as indicated in the second request, the query was subsequently forwarded to Agent 13. Ultimately, in the fourth request, Agent 0 successfully addressed the query with the corresponding data.

4.2 Discussion

This section delves into the performance aspects of our MASCD framework, exploring its response time, effectiveness, and overall efficiency. It is suggested that, for the sake of efficiency, we take the examples and use them for mapping out our framework. In the first part of this discussion, we will go over the factors using the MASCD Framework. For the first case study, we partitioned the database according to the mobile names and effectively distributed the data to each agent to prevent redundant data entry. Agents are connected cyclically in the approach that we have proposed. We accomplished this by utilizing DHT, which remembers the addresses of the nodes that come before and after it. It is a decentralized storage system that swiftly and efficiently looks through a vast volume of data to find information. We can find the data quickly and easily with the assistance of the DHT whenever a query needs to be run to locate it. Communication within multi-agent system is primarily determined by the degree of temporal complexity. Next, we will discuss the time required for our MASCD framework to reach completion. We have different mobile data in our database. We went into detail about it every time a user inquired about some mobile data and how it could be handled within the MASCD Framework. Figure 5 shows what happens when a user requests data about galaxy A04 from node 0 and how we resolve a query in an efficient time.

Figure 6.

Query response time.

Compared to alternative methods [2], MASCD responds to the query with half as many agents in the worst case. The worst-case time complexity of various approaches that don’t use a distributed hash table is O (n), as agents are unaware of which other agents have data or don’t. In order to locate data that is incompatible with the system, the query has to be sent to all of the other agents. Compared to other methods, the time it takes to find data using MASCD is significantly lower.

The effectiveness of the MASCD framework and other approaches is depicted in Fig. 6. The total number of agents is shown on the y-axis, while different strategies are shown on the x-axis. The findings show that MASCD is superior to other strategies. In [2] the technique of temporal massage decreases data redundancy in agent communication. Buffers are used for this purpose. The first is the sender buffer, while the second is the receiver buffer. Both the sender and the receiver may utilize buffers to hold messages temporarily. This technique likewise includes a message encoder and a local action generator, but the temporal complexity of finding the message increases by a factor of O (n) in the event of transmission loss. Message Dropout is also a technique for reducing data redundancy at the cost of communication overhead. In the worst situation, we have to move n/2 agents in MASCD. Alternative techniques need us to travel to each node in order to find the results. The primary issue in a multi-agent system is redundant data, which our MASCD framework addresses effectively and efficiently. Our architecture is effective if the number of products is greater than or equal to the number of agents. Otherwise, our framework cannot respond to the user’s query efficiently.

4.3 Conclusion and future work

The Multi-Agent System for Commodity Data (MASCD) Framework is presented in this study to reduce duplicate data in multi-agent system communication. Our suggested approach demonstrated how we effectively disseminated the database’s data to each agent after segmenting it based on product names. If replicas of the database are created, leading to redundant data, MASCD is more efficient than the centralized approach. We employ a DHT as a powerful querying tool. We can cut the overall number of agents searching for data in half via multi-agent system communication. Each agent keeps track of the addresses of the nodes that come before and after it. We count the steps to establish whether the query is transmitted clockwise or anti-clockwise. Finally, we use case studies from the real world to show how our suggested technique might be applied. In the future, we would like to integrate the option of the locality of reference of agents while distributing the database among the agents.

Footnotes

Author’s Bios

Awais Qasim received the M.S. degree in Computer Science from Lahore University of Management Sciences (LUMS), Lahore, Pakistan, in 2011. After that, he worked as a Software Engineer in Industry and developed a number of iPhone and Android applications. He is working as Assistant Professor in the Computer Science Department, Government College University. His research interests include model checking, multi-agent systems, real-time systems, self-adaptive systems, and robotics for disaster resilience. ORCID ID: 0000-0001-8677-9569.

Arslan Ghouri received his MSc degree from Punjab University College of Information Technology (PUCIT), Lahore, Pakistan, in 2020, and the M.S. degree in Computer Science from Government College University (GCU), Lahore, Pakistan. Currently, he is working in the research lab of Dr. Awais Qasim in the area of multi-agent systems. Following this, he is working as a Software Engineer in the industry, contributing to the development of various web applications. LinkedIn: https://www.linkedin.com/in/arslan-ghouri-1702481ba/.

Adeel Munawar received the M.S. degree in computer science from Government College University (GCU), Lahore, Pakistan, in 2018. His research interests encompass multi-agent systems, Artificial Intelligence, real-time systems, machine learning, and Natural Language Processing. Presently, he serves as a senior lecturer in Computer Science at Lahore Garrison University, actively engaged in research activities to promote research advancements. His scholarly contributions have been featured in numerous reputable journals and international conferences. Additionally, he is currently pursuing a Ph.D. degree from Sirindhorn International Institute of Technology, Thammasat University, Thailand. ORCID ID: 0000-0003-3315-3348.

References

Sharma

Srinivasan

and Sampath Kumar

, A comparative analysis of centralized and decentralized multi-agent architecture for service restoration, in: 2016 IEEE Congress on Evolutionary Computation (CEC), 2016, pp. 311–318.

Labrou

Finin

and Peng

, Agent communication languages: The current landscape, IEEE Intelligent Systems and Their Applications 14(2) (1999), 45–52.

Finin

Fritzson

McKay

and McEntire

, KQML as an agent communication language, in: Proceedings of the Third International Conference on Information and Knowledge Management, 1994, pp. 456–463.

Zhang

S.Q.

Zhang

and Lin

, Succinct and robust multi-agent communication with temporal message control, Advances in Neural Information Processing Systems 33 (2020), 17271–17282.

Mayfield

Labrou

and Finin

, Evaluation of KQML as an agent communication language, in Intelligent Agents II Agent Theories, Architectures, and Languages: IJCAI’95 Workshop (ATAL) MontrÃ©al, Canada, August 19–20, 1995 Proceedings 2, Springer, 1996, pp. 347–360.

Qasim

Amin

Aziz

Z.U.H.

and Khalid

, Efficient performative actions for e-Commerce agents, Applied Computer Systems, 25(1) (2020), page range. De Gruyter.

Wang

Qiu

and Rabinovich

, Learning efficient multi-agent communication: An information bottleneck approach, in: International Conference on Machine Learning, PMLR, 2020, pp. 9908–9918.

Yan

Jouandeau

and Cherif

A.A.

, A survey and analysis of multi-robot coordination, International Journal of Advanced Robotic Systems 10(12) (2013), 399. SAGE Publications Sage UK: London, England.

Bendjima

and Feham

, Multi-agent system for reliable routing in WSN, in: 2015 Science and Information Conference (SAI), IEEE, 2015, pp. 1412–1419.

10.

Mayfield

Labrou

and Finin

, Evaluation of KQML as an agent communication language, in: International Workshop on Agent Theories, Architectures, and Languages, Springer, 1995, pp. 347–360.

11.

Chen

Yang

L.T.

Kwon

Zhou

and Jo

, Itinerary planning for energy-efficient agent communications in wireless sensor networks, IEEE Transactions on Vehicular Technology 60(7) (2011), 3290–3299. IEEE.

12.

Nguyen

T.D.

Nguyen

Pham

V.-N.

Huynh

L.N.

Hossain

M.D.

and Huh

E.-N.

, Modeling data redundancy and cost-aware task allocation in MEC-enabled Internet-of-Vehicles applications, IEEE Internet of Things Journal 8(3) (2020), 1687–1701. IEEE.

13.

Kim

Cho

and Sung

, Message-dropout: An efficient training method for multi-agent deep reinforcement learning, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33(1), 2019, pp. 6079–6086.

14.

Son

Kim

Kang

W.J.

Hostallero

D.E.

and Yi

, Qtran: Learning to factorize with transformation for cooperative multi-agent reinforcement learning, in: International Conference on Machine Learning, PMLR, 2019, pp. 5887–5896.

15.

Kim

Jung

Cho

and Sung

, A maximum mutual information framework for multi-agent reinforcement learning, arXiv preprint arXiv:2006.02732, 2020.

16.

Liang

Wei

Jin

Huang

Cai

and Hua

X.-S.

, MaCAR: Urban traffic light control via active multi-agent communication and action rectification, in: Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, 2021, pp. 2491–2497.

17.

Mezghani

and Pr Mahmoud Abdellaoui, WSN intelligent communication based on Khalimsky theory using multi-agent systems, in: 2015 SAI Intelligent Systems Conference (IntelliSys), IEEE, 2015, pp. 871–876.

18.

Shehory

and Sturm

, Evaluation of modeling techniques for agent-based systems, in: Proceedings of the Fifth International Conference on Autonomous Agents, 2001, pp. 624–631.

19.

and Han

Q.-L.

, Distributed formation control of networked multi-agent systems using a dynamic event-triggered communication mechanism, IEEE Transactions on Industrial Electronics 64(10) (2017), 8118–8127. IEEE.

20.

Escrivá Gregori

Palanca Cámara

and Bada

G.A.

, A jabber-based multi-agent system platform, in: Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multiagent Systems, 2006, pp. 1282–1284.

21.

Logenthiran

and Woo

W.L.

, Intelligent multi-agent system for smart home energy management, in: 2015 IEEE Innovative Smart Grid Technologies-Asia (ISGT ASIA), IEEE, 2015, pp. 1–6.

22.

Zhang

Zeadally

Wang

and Liu

, Cognitive radio spectrum sensing framework based on multi-agent architecture for 5G networks, IEEE Wireless Communications 22(6) (2015), 34–39. IEEE.

23.

Wang

and Yin

, Realizing railway cognitive radio: A reinforcement base-station multi-agent model, IEEE Transactions on Intelligent Transportation Systems 20(4) (2018), 1452–1467. IEEE.

24.

Hunt

Meng

Hinde

and Huang

, A consensus-based grouping algorithm for multi-agent cooperative task allocation with complex requirements, Cognitive Computation 6 (2014), 338–350. Springer.

25.

Z.-G.

and Huang

, Event-triggered control for consensus of multiagent systems with fixed/switching topologies, IEEE Transactions on Systems, Man, and Cybernetics: Systems 48(10) (2017), 1736–1746. IEEE.

26.

Z.-G.

Pan

Y.-J.

and Tang

, Event-triggered control for consensus problem in multi-agent systems with quantized relative state measurements and external disturbance, IEEE Transactions on Circuits and Systems I: Regular Papers 65(7) (2018), 2232–2242. IEEE.

27.

Liu

Soh

Y.C.

and Xie

, Event-triggered communication and data rate constraint for distributed optimization of multiagent systems, IEEE Transactions on Systems, Man, and Cybernetics: Systems 48(11) (2017), 1908–1919. IEEE.

28.

Meng

Wang

and Liu

, Distributed load sharing of an inverter-based microgrid with reduced communication, IEEE Transactions on Smart Grid 9(2) (2016), 1354–1364. IEEE.

29.

Liu

Wan

and Lewis

F.L.

, Adaptive optimal decision in multi-agent random switching systems, IEEE Control Systems Letters 4(2) (2019), 265–270. IEEE.

30.

Qasim

Kanwal

Khalid

Kazmi

S.A.R.

and Hassan

, Timed-arc petri-nets based agent communication for real-time multi-agent systems, International Journal of Advanced Computer Science and Applications 10(9) (2019), Science and Information (SAI) Organization Limited.

31.

Long

Liu

and Pan

, Deep-learned collision avoidance policy for distributed multiagent navigation, IEEE Robotics and Automation Letters 2(2) (2017), 656–663. IEEE.

32.

Chen

Deng

Hua

Duan

Wang

and Wu

, Communication-reducing algorithm of distributed least mean square algorithm with neighbor-partial diffusion, Circuits, Systems, and Signal Processing 39 (2020), 4416–4435. Springer.

33.

Pantelimon

Tepe

Carriveau

and Ahmed

, Survey of multi-agent communication strategies for information exchange and mission control of drone deployments, Journal of Intelligent & Robotic Systems 95 (2019), 779–788. Springer.

34.

Pflaum

Szabo

and Coyle

, Communication collective: an adaptive multiagent communication platform, in: 2019 Winter Simulation Conference (WSC), IEEE, 2019, pp. 690–701.

35.

Gokulan

B.P.

and Srinivasan

, Distributed geometric fuzzy multiagent urban traffic signal control, IEEE Transactions on Intelligent Transportation Systems 11(3) (2010), 714–727. IEEE.

36.

Martínez

D.A.

Mojica-Nava

Watson

and sländer

, Multiagent self-redundancy identification and tuned greedy-exploration, IEEE Transactions on Cybernetics 52(7) (2020), 5744–5755. IEEE.

37.

Boubiche

D.E.

Bilami

and Toral-Cruz

, Big data challenges and data aggregation strategies in wireless sensor networks, IEEE Access 6 (2018), 20558–20571. IEEE.

38.

Das

Gervet

Romoff

Batra

Parikh

Rabbat

and Pineau

, Tarmac: Targeted multi-agent communication, in: International Conference on Machine Learning, PMLR, 2019, pp. 1538–1546.

39.

Shalev-Shwartz

Shammah

and Shashua

, Safe, multi-agent, reinforcement learning for autonomous driving, arXiv preprint arXiv:1610.03295, 2016.

40.

Tanner

, Distributed hash tables in p2p systems-a literary survey, Seminar on Internetworking, 2005, Citeseer.

41.

Bhullar

R.K.

Pawar

Kumar

et al., A novel prime numbers based hashing technique for minimizing collisions, in: 2016 2nd International Conference on Next Generation Computing Technologies (NGCT), IEEE, 2016, pp. 522–527.

42.

Bellifemine

Poggi

and Rimassa

, Developing multi-agent systems with a FIPA-compliant agent framework, Software: Practice and Experience 31(2) (2001), 103–128. Wiley Online Library.

43.

Juneja

Jagga

and Singh

, A review of FIPA standardized agent communication language and interaction protocols, Journal of Network Communications and Emerging Technologies 5(2) (2015), 179–191.

44.

Wooldridge

, An introduction to multiagent systems, 2009, John Wiley & Sons.

An effective approach for reducing data redundancy in multi-agent system communication

Abstract

Keywords

1. Introduction

2. Literature review

3.1.1 Objective function

3.2.1 Directory facilitator (DF)

3.2.2 Agent management system (AMS)

3.2.3 Agents

3.2.4 Distributed hash table (DHT)

3.2.5 Finger Table (FT)

3.2.6 Agent communication channel (ACC)

3.2.7 Agent communication language (ACL)

3.3 Query in a distributed hash table (DHT)

Table 2 a-z prime alphabets

4.1 Case study of e-commerce agents using mobile phones

4.1.1 Calculating hash indexes

Table 4 Split database on the base of product name

4.2 Discussion

Footnotes

Author’s Bios

References

Table 2
a-z prime alphabets

Table 4
Split database on the base of product name