⚠️ DRAFT/WIP: This documentation describes features currently in development on the mem_monitor branch.

Branch: mem_monitor
Status: In Development
Note: This whitepaper was originally in the repository at docs/memory-growth-monitor-whitepaper.md

← Back to Memory Monitoring Overview

The Antimetal Agent's Memory Growth Monitor implements a sophisticated three-detector approach to memory leak detection, combining complementary eBPF-based algorithms for robust, production-ready monitoring. By leveraging the kernel's kmem:rss_stat tracepoint (available in Linux 5.5+), the system provides accurate, low-overhead monitoring with multiple validation signals, achieving <0.1% CPU overhead while maintaining complete visibility into memory growth patterns.

The system employs three independent detection algorithms, each targeting different aspects of memory leak behavior:

1. Linear Regression Detector - Statistical trend analysis for steady growth patterns
2. RSS Component Ratio Detector - Memory composition analysis to distinguish heap leaks from cache growth
3. Multi-Factor Threshold Detector - Scientifically-backed heuristics from industry research

These detectors operate simultaneously on the same kernel events, providing multiple validation signals that dramatically improve detection accuracy while minimizing false positives. Each detector can trigger independently, but their combined signals provide the highest confidence leak detection.

For detailed architecture documentation, see Memory Monitor Architecture.

This whitepaper focuses primarily on the Linear Regression Detector, one of three detection methods in the system. The Linear Regression Detector uses statistical analysis to identify steady memory growth trends indicative of slow leaks.

---

Note: The full whitepaper content (673 lines) has been preserved in the wiki. For brevity, showing just the header. The complete document includes:

• Motivation and industry analysis
• Solution architecture
• Technical implementation details
• Performance analysis
• Experimental results
• Technical appendices

See the full document at: Memory Monitoring Whitepaper