MemProfiler: A tracing and analysis system for memory fragmentation

Memory fragmentation is a critical challenge in software, causing allocation failures despite ample free memory. Existing profiling tools focus on memory leaks or peak usage but fail to explain why fragmentation occurs. In this paper, we present a non-intrusive tracing and analysis system designed for fragmentation diagnosis. The system captures cross-layer memory events and reconstructs heap layout snapshots through offline replay. By introducing peak-anchored spatio-temporal focusing, the system compresses massive event streams into high-information-density segments. The system aligns metric anomalies with spatial evidence and triggering behaviors, providing system researchers and performance engineers with a traceable, and reproducible diagnostic capability for fragmentation analysis.