METHOD FOR LOCAL PREDICTION OF CASCADING QUEUE OVERLOADS IN REAL-TIME OPERATING SYSTEMS

Abstract

The article addresses the problem of early detection of cascading message accumulation in sequences of tasks and queues in real-time operating systems. Unlike monitoring an individual queue, the proposed method constructs a local subgraph starting from the first hazardous queue, estimates message arrival and processing rates, predicts backlog propagation along the task–queue chain, and accounts for possible changes in the system operating mode. As a result, the method determines not only the expected residual processing delay, but also the probability that message handling will be completed before the specified deadline.

The method was evaluated using simulation scenarios with different numbers of interconnected tasks. Comparison with simple queue-fill threshold monitoring showed that the proposed approach detects future deadline violations more accurately and provides a larger time margin for corrective action. A two-level warning scheme makes it possible to generate an early risk signal separately from a critical alert, thereby achieving a better balance between missed hazardous events and false alarms. It also supports responses ranging from preventive load redistribution to immediate intervention when the predicted risk becomes unacceptable.

The study is limited by simulation-based validation and by the need to specify the maximum permissible message age. Nevertheless, the input features and decision criteria do not depend on the services of a particular kernel. Therefore, the method can be adapted for different real-time operating systems and for systems with varying task structures, queue capacities, and workload dynamics.