Alexander G. Feoktistov

Multiagent approach to controlling distributed computing in a cluster Grid system

Generally, distributed computing environments have a number of properties that significantly complicate unification of processes of computing control such as their scheduling and resource allocation. Such properties include, for instance, functional-organizing heterogeneity, dynamics and non-complete description of integrated resources; diversity of the spectrum of problems solved using these resources; different classes of users pursuing their own goals and tasks while working with the computing system. Analysis of world trends in this research domain allows stating that solving these problems is directly connected with intellectualization of middleware of distributed computing environments with decentralized control. In this work, a multiagent approach to controlling distributed computing in a cluster Grid system, which is a virtual software-hardware infrastructure with its nodes being computer clusters, is presented. Characteristics of such system are considered. Architecture and principles of operation of the multiagent system are given. A number of important technological features of the proposed approach is singled out. The multiagent system is developed using JADE tools (Java Agent DEvelopment framework). The results of imitation simulation of the processes of operation of the system of agents being developed are given.

Service-oriented multiagent control of distributed computations

Consideration was given to the multiagent methods and toolkits for efficient control of the job flows generated by the service-oriented applications. These designs were integrated within the framework of a unique technology supporting automation of solution of large scientific problems in the up-to-date cluster Grid whose computing nodes (clusters) can be of an involved hybrid structure. The novelty and practical significance of the methods and tools described in the paper lie in essential extension of the functionality of the computation control system of the cluster Grid, as compared with the existing ones, distribution and sharing of the Grid resources at various levels of job execution, and possibility of integrating intelligent computation control tools in the problem-oriented applications.

Multiagent control of computational systems on the basis of meta-monitoring and imitational simulation

This paper describes the control of computations in a distributed computing environment (DCE) on the basis of its meta-monitoring and simulation modeling. Computations are controlled by a multiagent system with a given organizational structure. Resource allocation is carried out by agents with the use of economic mechanisms for controlling their supply and demand. Controlling actions for agents are formed on the basis of the simulation modeling of functional processes of the DCE. Data about the DCE resources and processes are collected and emergency situations in the DCE nodes are detected and prevented by the meta-monitoring system of this environment. The research results are the techniques for selecting control actions and the methods for intellectual processing and effective storage of data.

Knowledge elicitation in multi-agent system for distributed computing management

The effective management of scalable applications for solving large problems in a heterogeneous distributed computing environment is the non-trivial problem. Scalable applications generate competitive job flows that have be executed with the help of shared resources of the environment. The promising approach to solve this problem is to use multi-agent technologies. To this end, we develop a multi-agent system for the management of scalable applications. In contrast to known multi-agent systems, our system is based on applying a special conceptual model of the environment. It includes several components of a comprehensive knowledge about both the environment and subject domains of solved problems. We propose a new approach to an elicitation of these knowledge components through an integrated use of the conceptual modelling of distributed computing, classification of jobs and resources, and parameters adjustment for agent algorithms. With this approach, specialists in various fields of distributed computing considered as users of the environment, can apply their own knowledge at different levels of the problem solving process. This flexibility and algorithm adaption are benefits of our approach. Extensive modeling and practical experiments, with variation of important parameters of applications execution, show the efficiency of our management under developed multi-agent system for scalable applications.

Automation of multi-agent control for complex dynamic systems in heterogeneous computational network

The rapid progress of high-performance computing entails new challenges related to solving large scientific problems for various subject domains in a heterogeneous distributed computing environment (e.g., a network, Grid system, or Cloud infrastructure). The specialists in the field of parallel and distributed computing give the special attention to a scalability of applications for problem solving. An effective management of the scalable application in the heterogeneous distributed computing environment is still a non-trivial issue. Control systems that operate in networks, especially relate to this issue. We propose a new approach to the multi-agent management for the scalable applications in the heterogeneous computational network. The fundamentals of our approach are the integrated use of conceptual programming, simulation modeling, network monitoring, multi-agent management, and service-oriented programming. We developed a special framework for an automation of the problem solving. Advantages of the proposed approach are demonstrated on the parametric synthesis example of the static linear regulator for complex dynamic systems. Benefits of the scalable application for solving this problem include automation of the multi-agent control for the systems in a parallel mode with various degrees of its detailed elaboration.The rapid progress of high-performance computing entails new challenges related to solving large scientific problems for various subject domains in a heterogeneous distributed computing environment (e.g., a network, Grid system, or Cloud infrastructure). The specialists in the field of parallel and distributed computing give the special attention to a scalability of applications for problem solving. An effective management of the scalable application in the heterogeneous distributed computing environment is still a non-trivial issue. Control systems that operate in networks, especially relate to this issue. We propose a new approach to the multi-agent management for the scalable applications in the heterogeneous computational network. The fundamentals of our approach are the integrated use of conceptual programming, simulation modeling, network monitoring, multi-agent management, and service-oriented programming. We developed a special framework for an automation of the problem solving. Advantages of the ...

The Service-Oriented Multiagent Approach to High-Performance Scientific Computing

The tools for intelligent management of high-performance computing in a heterogeneous distributed computing environment for solving large scientific problems are represented and the service-oriented multiagent approach to solve such problems using these tools is proposed. A purpose of our research is expansion of opportunities for management of the considered environment. Advantages of the proposed approach as compared with approaches based on use of the traditional systems for a distributed computing management are illustrated with two examples of scientific services. Experimental results show a high scalability and efficiency for calculations carried out with use of these services.

Scalable Application for the Search of Global Minima of Multiextremal Functions

This paper describes the urgent issue of providing scalability of computations in the solution of multiextremal problems arising in different fields of scientific studies, including image processing. There is an approach proposed for the development of the Gradient scalable application for solving the problem of global optimization of multiextremal functions with account for a multistart method in the Orlando framework. An additional step of computations is implemented in the problem solving scheme, which makes it possible to decompose the problem with account for the performance of computational resources and thereby minimize the time it takes to solve it as opposed to a classical multistart method. Special agents of the metamonitoring system for measuring the performance of resource with regard to the problem solved are developed.