Explore the challenges and solutions in performance tuning of microservices in data centers through this conference talk. Dive into the complexities of optimizing multiple microservices with varying workloads and numerous configuration options. Learn how Bayesian optimization-based machine learning can be applied to tackle this combinatorially intractable problem. Discover the pitfalls and lessons learned from implementing a continuous optimization service for microservices. Gain insights into maintaining optimal performance despite ongoing upgrades to service, platform software, and hardware. Understand the potential for improving resource utilization and unlocking hidden performance gains in data centers. Follow along as the speaker, a Staff Engineer in Platform Engineering at Twitter, shares experiences and outlines a vision for a continuous optimization service in microservice-based architectures.
Continuous Optimization of Microservices Using Machine Learning
Overview
Syllabus
Introduction
The Problem
Performance Stack
Performance Tuning
Performance Optimization
Performance Constraints
Hidden Variables
Performance Tuning Problem
Bayesian Optimization
Example
Gaussian Process
Expected Improvement
Bayesian Optimization as a Service
Bayesian Optimization API
Random Search
Twitter
Recap
Microservice
Staging
Setup
Results
Optimization changes
Takeaways
Implementation
Conclusion
Question
Taught by
Devoxx
