On some approaches to increase the fault tolerance of the computing cluster of an intelligent transport system, taking into account the priorities of the tasks

Abstract

Background. The article is devoted to the study of methods for ensuring fault tolerance of cluster computing systems used in intelligent transport systems (ITS). Approaches to the rational allocation and redistribution of tasks in the conditions of failures of computing modules, as well as issues of ensuring the required level of performance with limited resources and an increasing probability of failures are considered. Materials and methods. The paper analyzes two classes of methods: static fault-tolerant task reallocation and dynamic priority-based task allocation. An algorithm based on sorting tasks by weights and selecting the least loaded nodes is proposed for the dynamic scheduler. The fault tolerance testing procedure is described separately, including step-by-step simulation of failures with task execution time control. Results. Experiments conducted on a cluster system of 8 nodes and three sets of tasks have shown that the developed dynamic algorithm provides an average of 38 % shorter execution time compared to a random distribution. In case of node failures, the system using the developed approach retains the ability to perform a greater number of priority tasks within the acceptable time. Conclusions. The developed algorithm demonstrates efficiency for clusters with a fixed set of independent tasks and strict time constraints typical for ITS. It increases fault tolerance and ensures more efficient use of resources compared to universal orchestration systems, offering a specialized solution for real-time data processing tasks.