Predictive Reliability Assessment Model for Containerized Enterprise Software Systems

Abstract

Containerized enterprise software systems require early reliability assessment because failures can develop through resource pressure, repeated restarts, slow dependencies, and unstable orchestration. This study presents a predictive reliability model that combines container restart behaviour, processor and memory pressure, service-response degradation, dependency health, and orchestration instability. Runtime metrics were collected across fixed monitoring windows and converted into temporal prediction features. The model was evaluated under six increasing failure-stress levels and compared with threshold-based monitoring and history-only assessment. Results showed that the proposed approach followed the observed system condition more closely and identified reliability decline earlier than the comparison methods. It also provided indicator-level warnings that explained the main causes of reliability loss. The framework can support proactive maintenance, faster operational decisions, and reduced service interruption by helping enterprise teams detect developing container failures before they become severe and affect dependent services across distributed production environments during demanding and rapidly changing workloads.