Java at Speed: Building a Better JVM
summary
Zing replaces HotSpot components to deliver concrete results: the C4 garbage collector runs fully concurrent and compacting to eliminate GC pauses (claims scale to ~20 TB heaps without longer pauses), the Falcon JIT embeds LLVM to emit AVX2 vectorized code for much higher throughput on numeric loops, ReadyNow! loads profiles and compiled code so restarts hit ~98% of peak after two or three transactions, and jHiccup measures JVM latency for apples‑to‑apples comparisons on Linux and container heaps under 1 TB.
AI-written summary · about
chapters (25)
- 0:00Intro
- 1:10JVM Performance Graph: Ideal
- 1:30JVM Performance Graph: Reality
- 5:18What If There Was A Better JVM?
- 5:36Azul Zing JVM
- 7:11Zing System Tools
- 9:40C4 Basics
- 11:40Loaded Value Barrier
- 14:26Concurrent Mark Phase
- 16:04Relocation Phase
- 17:03Remapping Phase
- 18:03Zing: Big Heaps, No Problem
- 18:52GC Tuning Used To Be Hard
- 19:48GC Tuning With Zing
- 20:21Measuring Platform Performance
- 21:22Big Heap, Small Latency
- 22:48Small Heap, Small Latency
- 23:21Advancing Adaptive Compilation
- 24:11Simple Code Example
- 24:49More Complex Code Example
- 25:09Traditional JVM JIT
- 25:34Falcon JIT
- 26:57ReadyNow! Warmup Elimination • Save JVM JIT profiling information - Classes loaded
- 28:52Effect Of ReadyNow!
- 29:25The Zing JVM
related talks
44mJava at Speed: Building a Better JVM.
51mJava at Speed: Getting the Most Out of Modern Hardware
01h 03mKeeping Your Java Hot.
48mJava 9 Performance
01h 01mAn Introduction to JVM Performance
48mJust-In-Time Compilation for Java Performance: Recent and Ongoing Improvements
11mJVM internals — bytecode, JIT, GC, speculative optimization
01h 32mUnderstanding Java Garbage Collection and what you can do about it