Performance Tuning Twitter Services with Graal and Machine Learning

Explore performance tuning techniques for Twitter services using Graal and Machine Learning in this 52-minute Devoxx conference talk. Dive into the successful implementation of Graal at Twitter, which has significantly reduced datacenter costs. Learn about the Autotune Machine Learning framework and its application in optimizing Graal inlining parameters. Discover the principles of Bayesian optimization and its role in finding optimal performance settings. Examine real-world experiments, including PS scavenge cycles, user CPU time, and latency improvements. Analyze detailed results through tables, charts, and low-level graphs. Gain insights into Twitter's quest for the "Holy Grail" of performance optimization, and understand how Autotune parameters impact inlining and overall system efficiency. Witness the practical applications of these techniques across various Twitter services, including Social Graph and Orange Control experiments.

Intro
Performance tuning
Performance tuning internal
Bayesian optimization
How Bayesian optimization works
What does it do
How it finds
Optimality
Autotune
What is Autotune
What is Graal
Open JDK
Inlining parameters
Twitters Quest for Holy Grail
PS scavenge cycles
User CPU time
Ranges
Test setup
Objective
Constraints
Experiments
Results
Results Table
Results Chart
Maximum Landing Site
Low Level Graph
Verification Experiment
Data Visualization
CPU Time
Latency
Performance improvements
Experiment 2 Social Graph
Experiment 3 Social Graph
Experiment 4 Orange Control
Experiment 4 Results
Verification Run
Social Graph
Autotune Social Graph
Autotune parameters
Inlining
Evaluation
Outcome
Max in line size
Inline small code
Autotuned