Migrating the Jira and Confluence applications to AWS Graviton

Summary

The article outlines Atlassian's migration of over 3,000 Jira and Confluence instances to AWS Graviton, emphasizing the performance and cost benefits of the transition. It details the technical challenges encountered, including latency issues and CPU performance discrepancies between Graviton and Intel architectures. The authors discuss their methodology for benchmarking performance, highlighting the importance of identifying system bottlenecks before conducting micro-benchmarks. They also explore the significance of CPU cache management and the impact of TLB pressure on performance, providing insights into the optimizations made to improve efficiency during the migration process.

Key Learnings

• 1Understanding the performance trade-offs between different CPU architectures is crucial for optimizing application efficiency.
• 2Identifying system bottlenecks at a high level before micro-benchmarking can lead to more effective performance improvements.
• 3Effective CPU cache management and TLB optimization are essential for maximizing application performance on new architectures.
• 4Collaborative agreement on benchmarking methodologies can streamline performance testing and yield more reliable results.
• 5Addressing capacity issues, such as ICE errors, is vital for successful scaling in cloud environments.

Who Should Read This

Senior Site Reliability Engineers and Performance Engineers optimizing large-scale Java applications on AWS infrastructure.

Test Your Knowledge

Topics