JVM Performance Engineering • Monica Beckwith & Kirk Pepperdine • GOTO 2024

Performance engineering requires understanding the “what”, “why” and “how” - focusing on both the high-level system behavior and low-level implementation details

Experimental design and benchmarking are core activities, but most benchmarks are flawed due to complexity of variables and difficulty in reproducing real-world scenarios

Top-down and bottom-up methodologies are complementary approaches - top-down starts with system-level analysis while bottom-up focuses on specific optimization areas

Hardware knowledge is critical for performance engineering, as JVM performance is heavily influenced by CPU architecture, memory systems, and hardware capabilities

Observability is essential for performance engineering, providing insights into system behavior and enabling data-driven decisions

Non-functional requirements (“ilities”) like scalability, reliability, availability and maintainability are key aspects of performance engineering

Performance engineering should be integrated early in the development lifecycle, not just during deployment

Modern challenges include cloud native environments, microservices, containers which add complexity to performance optimization

Collaboration across disciplines (hardware, software, operations) is necessary as systems become more complex

The JVM continues evolving to handle new hardware capabilities, concurrency models, and deployment scenarios requiring ongoing performance optimization work