**4. Conclusion**

overloaded resources (sources) and the underloaded ones (receivers), depending on their load information by using the threshold values. Grid Agent and Cluster Agent

> **Location policy**

receiver resource is the most under loaded and it has the highest communication load with the migrated Agent

receiver resource is the least loaded host

The destination node is the node with the least LC (Location Credit) value

having desired configuration

available worker agent in **Decision making**

Migration decision is taken by Cluster Agent

Migration decision is taken by migration management agent

Migration decision is taken locally by the LBC Agent

Migration decision is taken by Load agent

Migration decision is taken by

**Migration condition**

Cluster state is unbalanced;

host is overloaded,

Local load value is greater than the load threshold value

fitness value of a VM becomes less than or equal to threshold value

network traffic analysis

**Implementation**

Jade [6] + Alea2 [7] simulator

AST-RTI [11] version 2.0 + C+ +

Java + Jade

Java

Jade

interfaces are shown in **Figures 5** and **6**.

**3.3 Model comparison with some works**

**Load information gathering policy**

Event-based information gathering

periodicbased information gathering

Event-based information gathering

periodicbased information gathering

periodicbased information gathering

**Selection policy**

*Artificial Intelligence - Latest Advances, New Paradigms and Novel Applications*

Credit -based concept

computation and communication cost

Credit -based concept

Not cited receiver VM

task cost receiver is

**configuration**

contains a set of computing resources hierarchy of control, with six types of agents

The system contains entity, federate, VM, and host with migration management agent

The system contains a set of nodes decentralized in control, with seven agents

contains a set of VMs centralized in control, with three agents

The system contains a set of machines centralized in

A2LB [12] The system

N LB WITH STRONG MIGR ATION IN AN

**68**

**Model System**

**Figure 6.**

*Cluster agent interface.*

ABLBM The system

VM, dynamic balancing [10]

LB in distributed MAS [4]

Recognizing key factors to prove the convergence of grid and MAS and models is not a simple task. We note that the current state of GRID and MAS research activities are necessarily developed to enable justifying the study of the path towards an integration of the two fields.

We have presented a theoretical comparison between some related works and the proposed model. The proposed model has some unique features. It is hierarchical, which facilitates the circulation of information through the tree and defines the flow of messages between agents. Also, the proposed Agent-based load balancing model uses an event-driven information gathering policy, the latter being especially beneficial in terms of economy of usage of network resources. Furthermore, it can achieve excellent performance with significantly less computational load and system instability than a periodic information gathering policy. To select the migrating agent, we use the credit-based concept, accordingly, some factors are considered to calculate the credit value. Moreover, in the selection of receiver resources, we take into consideration the resource loads and the communication between the receiver resources and the migrating agent for avoiding the migration for external resources and reducing the communication cost. The migration decision is taken locally by Cluster Agent, where each cluster agent to balance its load among its associated resources. If it fails, the Cluster Agent migrates worker agents to underloaded clusters based on the load information received by other clusters. Finally, it supports flexibility and expandability, thus, various intelligent agents have been deployed to decrease system complexity by modularization. Moreover, it is easy to modify its components, and add more features and functions to it.

In theory, the multi-agent architecture of load balancing systems introduces important improvements, such as better average performance when one computer is not working and a lower system-error probability. In terms of the development process, fault-tolerance, and scalability, the agent approach offered the expected improvements, both in objective real-world measurements and in the subjective observations of designers, developers, and users.

On another hand, we could not overcome several well-known problems when designing distributed systems. For example, handling failed entities, synchronization problems, and query-response-related issues turned out to be the same as in any distributed programming. It is important to be aware of the advantages and disadvantages of the agent and non-agent approaches, but the most important point is whether the advantages prevail. For load balancing systems, our theoretical analysis and practical experiences both indicate that the advantages of agent-based load balancing systems clearly be more than the observed disadvantages.

The system performance was not studied yet. Thus, there is a need to analyze execution efficiency and compare it to available Agent-based load balancing platform evaluations. Further research is going to concentrate on execution performance.

*Artificial Intelligence - Latest Advances, New Paradigms and Novel Applications*

**References**

ADCAIJ201984518

MGS-200326.

[1] A. Wided, K. Okba, and B. Fatima. Load balancing with Job Migration Algorithm for improving performance on grid computing: Experimental Results. Advances in Distributed Computing and Artificial Intelligence Journal, 2019; 8(4): 5–18. DOI:10.14201/

*Agent Based Load Balancing in Grid Computing DOI: http://dx.doi.org/10.5772/intechopen.94219*

(CCPE),2002, 14:13–15. DOI: 10.1002/

[9] N. Li, X.-Y. Peng, M.-H. Zhang, M. Wang, and G.-H. Gong, Multimedia communication over HLA/RTI, SimulationModelling Practice and Theory,2006,14(2): 161–176. https://doi. org/10.1016/j.simpat.2005.03.003

[10] A.Singh,D. Juneja, M.Malhotra. Autonomous agent based load balancing

[11] X. Song, Y. Ma, D.Teng. A load balancing scheme using federate

cloud simulations, Mathematical Problems in Engineering, 2015,pp.1–11. https://doi.org/10.1155/2015/506432

an agent based grid using CSP

Computing & Applications (IJGCA),2012, 3(4),43–53. DOI:

10.5121/ijgca.2012.3404

workload/l\_metacentrum

migration based on virtual machines for

[12] Z. A .Sayar and N.Erdogan. Network load balancing with strong migration in

Approach. International Journal of Grid

[13] http://www.cs.huji.ac.il/labs/parallel/

International Conference on Advanced

algorithm in cloud computing.

Computing Technologies and Applications, 2015. DOI: 10.1016/j.

procs.2015.03.168

cpe.710

[2] A. Wided, K. Okba . A new agent based load balancing model for improving the grid performance. Multiagent and Grid Systems Journal, 2020; 16(2):153–170. DOI: 10.3233/

[3] Chow K-P, Kwok Y-K. On load balancing for distributed multiagent computing. IEEE Trans Parallel Distrib

[4] A.M. Metawei, G. Salma,M.H. Sahar ,

Syst ,2002;13(8):1153–61.

M.N.Salwa .'Load balancing in distributed multi-agent computing systems', Ain Shams Engineering Journal,2012, 3(3):237–249.

[5] R. Gupta . Coding categorical variables in regression models: dummy and effect coding.Cornell Statistical Consulting Unit,2008,4(2):202–210.

[6] F.L. Bellifemine, G.Caire, and D. Greenwood, Developing Multi-Agent Systems with JADE. John Wi-ley & Sons,2007, NJ. ISBN: 978–0–470-05747-6

[7] D. Klusáček and H.Rudová. Alea 2 job scheduling simulator. In Proceedings of the 3rd International Conference on Simulation Tools and Techniques (SIMUTools 2010), Torremolinos, MalaMA, Spain. DOI: 10.4108/ICST.

[8] R.Buyya, and M.Murshed. Gridsim: a toolkit for the modeling and simulation of distributed resource management and scheduling for grid computing, The

Computation: Practice and Experience

SIMUTOOLS2010.8722

Journal of Concurrency and

**71**
