Finalization-Based Object Lifetime Profiling

on YouTube ↗15m

" summary

FiLiP instruments every allocation, attaches an ephemeron to record finalization time, and supports sampling (100, 50, 1, 0.1%) and optional pretenuring ephemerons. Measurements show usable object lifetime estimates even with 0.1% sampling, and pretenuring reduces profiler overhead by 3.68× on average with an observed peak improvement of ~15× on one benchmark. The implementation is practical on existing VMs using ephemeron finalization and yields actionable lifetime data with much lower overhead.

AI-written summary · about

" chapters (7)

1. 0:00Finalization-based profiling intro
2. 1:50Ephemeron-based profiling
3. 3:55Research questions & benchmarks
4. 6:03Sampling impact on precision
5. 9:37Execution overhead analysis
6. 11:27Summary of findings
7. 12:10Q&A