PREDICTING THE DEADLOCK STATE OF PROCESSES IN COMPUTER SYSTEMS

Abstract

The article considers the problem of predicting the deadlock state of processes in computer systems operating under conditions of parallel process execution, dynamic resource allocation, and competitive access to memory, files, input/output devices, synchronization objects, transactions, and database records. The relevance of the study is determined by the fact that deadlock is a critical state in which several processes hold part of the required resources and simultaneously wait for resources occupied by other processes. This leads to reduced performance, disruption of service availability, and the impossibility of further execution of blocked processes without external intervention. It is shown that classical approaches to deadlock prevention, avoidance, detection, and recovery have certain limitations, since they often require prior knowledge of future process requests or respond only after a dangerous state has already been formed. A method for predicting the deadlock state of processes is proposed. It combine the formation of a process signature, a model of the current state of the computer system, automated construction of a fuzzy rule base, and assessment of the risk of process transition into a boundary or deadlocked state. The process signature is represented as a structured description of identification, temporal, resource, and behavioral characteristics. This makes it possible to take into account not only the fact of waiting for a resource, but also the context of the process interaction with other processes and system resources. Fuzzy C-Means clustering is used to form the rule base, which ensures the identification of typical resource states and takes into account the gradual transition between normal execution, regular waiting, boundary state, and deadlock. The proposed method is activated at the moments when a requested resource is not granted to a process. This reduces computational costs and focuses the analysis on potentially dangerous situations. The obtained results create a basis for preventive resource management in a computer system, early detection of deadlock prerequisites, and improvement of the stability of the computing environment.