The scheduler for the grid­system based on the parameters monitoring of the computer components

Abstract

The structure of the centralized distributed computer system (DCS) task scheduler, which uses the adaptive resource security management mechanism was developed. By interacting with local agents of the given compute nodes, the scheduler defines the system node parameters and selects the resources with the specified security and performance requirements. Ensuring an optimum combination of mutually exclusive security and performance parameters is a non-trivial task, requiring the development of new approaches to solving it. The research found that the adaptive distributed system resource security management mechanism increases the DCS performance in comparison with the classical resource security management mechanism. In particular, the research shows that the average task time in the queue and the average task time in the system with the adaptive security level management mechanism is 2.8 and 2.1 times lower, respectively, in comparison with the classical security level management mechanism. At the same time, the adaptive security management introduction requires additional software on the DCS compute nodes for the CN status parameters monitoring. The experiments demonstrate that the monitoring system can significantly reduce the DCS performance. Thus, according to the experiments, in case of 25 % load on the DCS CN from the monitoring system, the average task time in the queue and the average task time in the system increase by 62 % compared with a situation where monitoring is not performed. The research results need to be considered when introducing the secure data processing mechanisms in DCS to prevent a substantial decrease in the distributed system performance.