KERNEL MODULE LOAD

MONITORING LAYER

• Iterate over all runnable processes - Voluntary context switches

• Collect metrics (no decisions here): - Involuntary context switches

- CPU usage % per process

- Running time vs Waiting time

• Store metrics in a Process Metrics Structure

Output: Raw, unclassified process data

DETECTION LAYER

• Read Process Metrics Structure • Classify processes:

• Apply dynamic, rule-based thresholds: - Normal

- High CPU % for long duration - Resource-heavy

- Low waiting time + excessive run time - Responsiveness-degrading

- Abnormal context switch patterns

Output: Classified process list + severity level

CONTROL ACTION LAYER

• Receive classified processes - Reduce scheduling priority

• Decide corrective action WITHOUT killing process - Throttle CPU time

Possible actions: - Adjust scheduling class

- Temporarily deprioritize execution

• Apply action to offending process

SYSTEM RESPONSIVENESS RESTORED

(Processes continue, but under controlled use)