Dive into an in-depth 43-minute Java conference talk on JVM benchmarking presented by Aleksey Shipilev. Explore essential concepts including naive questioning, proper unit usage, and treating benchmarks as experiments. Learn about experimental control, common wisdom, root causes, and evil optimizations. Gain insights into JMH API, warmup plateaus, power management, and OS schedulers. Examine various optimization techniques such as dead-code elimination, constant folding, and profile feedback. Understand the intricacies of inlining budgets, cache capacity, bulk method transfers, and branch prediction. Conclude with a comprehensive understanding of the seriousness of benchmarking and future considerations in this field.
JVM Benchmarking Techniques and Pitfalls - Lecture by Aleksey Shipilev
Overview
Syllabus
Intro: Speaker's Credentials
Basics: Naive question
Basics: Getting the units right
Basics: Benchmarks are experiments
Basics: Experimental Control
Basics: Common Wisdom
Basics: The Root Cause
Basics: Evil Optimizations
Basics: Know Thy Optimizations
Basics: JMH API Sneak Peek
Basics: Warmup plateaus
System: Optimization Quiz (A)
System: Power management
System: OS Schedulers
VM: Optimization Quiz (B)
VM: Dead-code elimination
VM: DCE, Avoiding
VM: DCE, Blackholes
VM: Optimization Quiz (C)
VM: Constant folding, etc.
VM: DCE and CSE are brothers
VM: Optimization Quiz (D), #2
VM: Optimization Quiz (E). #3
VM: Profile feedback
VM: Optimization Quiz (F). #2
VM: Run-to-run variance
VM: Inlining budgets
VM: Inlining example
CPU: Cache capacity
CPU: Optimization Quiz (G) Scalability for this code?
CPU: Optimization Quiz (G). #2
CPU: Bulk method transfers
CPU: Optimization Quiz (H)
CPU: Branch Prediction
Conclusion: Benchmarking is serious
Conclusion: Things on list to do
Taught by
Java
