AOT or JIT: Faster Startup or Faster Code?
summary
Compares concrete AOT and JIT tradeoffs for JVM microservices: GraalVM native-image yields much faster startup and smaller memory (Oracle claims ~50× startup, ~5× smaller heap) but loses dynamic speculative inlining; JIT delivers most throughput via speculative optimizations (over half of gains) and adaptive recompilation. Describes Azul tech: C4 GC (continuous concurrent compacting, pauseless), Falcon (LLVM-based JIT), and ReadyNow + compile stashing (records classes, initialization, profiles, deopt info and cached machine code), cutting compile time ~80% and CPU load ~60%, and CRaC checkpoints that restore Spring Boot from ~4s to ~38ms.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:12JVM Performance Challenges
- 5:35Microservice-Based Application
- 9:21Microservice Performance Challenges
- 12:30Java Compilers: Ahead of Time (AOT)
- 17:46Compilers: Just in Time (JIT)
- 20:12Speculative Optimisations
- 27:09Graal VM For Java Applications
- 28:54OpenJDK Project Leyden
- 30:30The Azul Prime JVM
- 33:14ReadyNow! Solution
- 37:10ReadyNow! Effect
- 37:46ReadyNow! Start-up Time
- 38:32Falcon Pipeline
- 39:10Deterministic Compiler
- 39:59Add Compile Stashing
- 40:32Compile Stashing Effect
- 40:59Co-ordinated Restore at Checkpoint (CRC)
- 44:11Faster Startup or Faster Code
related talks
38mAOT or JIT: Faster Startup or Faster Code?
27mHow to Eliminate Java Performance Warmup
55mJIT and AOT in the JVM
01h 03mKeeping Your Java Hot.
51mAOT: Everything you need to know about GraalVM for Java applications
51mGoing AOT: Everything you need to know about GraalVM for Java applications
54mKeeping Your Java Hot by Solving the JVM Warmup Problem
50mGoing AOT: Everything you need to know about GraalVM for Java