GC Algorithms
summary
Compares garbage collection designs: reference counting (fast predictable pauses but needs cycle collection) versus tracing collectors that use mark-compact or three-space copying (reduces fragmentation but copying can waste over half the heap). Covers generational collectors with promotion heuristics, card tables for cross-generation pointers, per-thread allocation/promotion buffers, trade-offs of parallel vs concurrent collectors, and a metronome-style real-time mode that enforces fixed run/pause intervals (e.g. 5s run / 5s pause) at the cost of high scheduling overhead.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 4:23GC basics
- 6:57Reference counting
- 11:55Tracing vs ref counting
- 12:19Mark-sweep and mark-compact
- 15:33Choosing GC for app
- 17:01Copying GC benefits
- 21:02Generational hypothesis
- 23:56Generational collectors
- 26:22Parallel and concurrent GCs
- 28:10Concurrent GC benefits
- 31:40Real-time GC uses
related talks
46mGraalVM, CRaC, Leyden and friends - in search of TRULY...
11mRAGTrace: Understanding and Refining Retrieval-Generation Dynamics in Retrieval-Augmented Generation
53mThe Best We Can Do?
01h 09mRecent Concurrency and Parallelism Proposals to the C++ Standard Committee
15mGraphRAG methods to create optimized LLM context windows for Retrieval
28mOptimizing Context Tuning for RAG Systems
12mStriaTrace: Efficient Tracing and Diagnosis for Online LLM Inference
14mSmarter RAG, Smaller Bill: Optimize for Performance and Price