The work is devoted to the research of decentralized management of distributed computer system (DCS) resources. Decentralized resource management involves the presence of several elements of the system that are responsible for managing resources across the DCS and are equal to each other. With decentralized resource management, such processes as resource search, task scheduling, and others that are shared between DCS components, so if one of the components does not work, then the other takes on its functions. DCS with decentralized resource management are highly scalable. Nevertheless, the decentralization of management of distributed computing resources leads to serious problems in resource search, task scheduling, allocation of tasks to resources, coordination of tasks completion, and so on. The effectiveness of a DCS depends on the coordination and interaction between its elements. That is why a networkcentric approach has been chosen to improve the efficiency of managing the computing resources of a DCS.
As a result of the research, the method of managing the resources of the DCS has been improved, which, unlike existing ones, is based on a networkcentric approach, which made it possible to increase the efficiency of allocation and management of computing resources, taking into account the following parameters of the DCS functioning: system response time, waiting time. For resource selection, a comprehensive analytical model for determining the marks of the DCS's functioning was also developed. The proposed model allows us to evaluate the four marks of the DCS’s functioning, namely performance, data transfer speed, reliability and security, depending on the number of nodes of the DCS.
Furthermore, the method of determining the required amount of reserve resources for a task with an asynchronous mode of execution, which is based on the use of the neural network, has also been developed in the work, which also allowed to improve the efficiency of the DCS's functioning with the mechanism of resource management based on a networkcentric approach.
On the basis of the proposed models and methods, the tools for managing the computing resources of the DCS were developed, namely: a metascheduler of a DCS with support of secure data processing, a DCS resource management system module of a decentralized type on the base of communities of computing resources, a DCS resource management system module of hierarchical type, a DCS resource management system module of a decentralized type on the base of a network-centric approach. Secure data processing involves monitoring the security level of the DCS computing node, therefore, in this work a local data agent is also proposed for monitoring the performance, reliability and security of the computing node, as well as the speed of data transmission to and from it.
Experimental studies confirmed the effectiveness of the proposed models and methods. To carry out all the experiments, a special program complex was developed that implements simulation of distributed computer system functioning.
The main result of this research is to ensure that the system's response time is reduced by using a decentralized resource management mechanism based on a network-centric approach and reducing the number of exchanges by service messages. Thus, the use of a resource management mechanism based on a networkcentric approach has allowed us to reduce almost three times the number of service messages per task compared to the mechanism based on the formation of communities of computing resources.The task waiting time in the system with the mechanism of resource management based on the network-centric approach decreased almost twice, compared with the mechanism based on the formation of computing communities, while the time of execution of the task package decreased by only 16%-20%. This is due to the fact that the time of the tasks completion has increased. The reason for this is the emergence of situations of lack of free resources with the necessary productivity or lack of free resources in general.The practical value of the work is to develop a software package that allows you to simulate the work of the DCS and explore its functioning with the application of various distribution mechanisms and resource management. The software package supports synchronous and asynchronous modes of the task execution,simulates classical and adaptive mechanisms for controlling the security of a computing node, models the presence or absence of a computing node monitoring system, supports various mechanisms of distribution and management of resources (the mechanism of centralized management of resources with secure data processing, the mechanism of hierarchical control and resource allocation, the mechanism of decentralized resource management based on communities of computing resources, the mechanism of decentralized resource management based on networkcentric approach), performs collection of statistical data
