Monitoring collects and analyzes measurements over time to understand system behavior, compare results with requirements, and detect changes or problems. Good monitoring starts by defining what “good” means for the situation: latency may matter most for patient monitoring, while concurrent linked processes may matter most in a chemical plant. section 7.1
Use monitoring as a practical feedback loop:
- Define relevant targets and requirements before judging performance. section 7.1
- Collect measurements such as execution time, memory allocation, response speed, accuracy, and relevance. section 7.1 section 8.2.2
- Compare the current system with earlier versions or expected behavior after changes. section 7.1
- Review results regularly, investigate misleading or incorrect outputs, and intervene before small issues become larger failures. section 8.2.2
- Record use cases, data sources, preprocessing, reviews, and performance findings so monitoring supports accountability rather than box-ticking. section 8.2.2
From infrastructure change to useful signals
After a cloud migration makes processing dramatically slower, profiling can reveal which functions or algorithms consume the most time or memory. Stress testing then examines behavior near or beyond expected loads. For an AI model, monitoring also checks whether responses remain accurate and relevant, not merely fast. section 7.1 section 8.2.2
