Benchmarking Amazon Aurora Limitless with pgbench

This article provides a comprehensive guide to benchmarking Amazon Aurora Limitless using pgbench, highlighting the key considerations and best practices for testing performance in a distributed database environment.

• Key Benchmarking Considerations:
  • Sharding strategy and shard key selection
  • Table collocation and distribution
  • Workload optimization for multi-node systems
  • Importance of including shard key predicates in queries
• Performance Improvements:
  • Initial run showed 518 TPS, improved to 1,538 TPS with workload modifications
  • Extended 1-hour test reached over 2,000 TPS
  • Concurrent router testing achieved over 7,000 TPS
  • Extreme test with high-capacity setup reached 2.5M TPS
• Key Recommendations:
  • Choose shard keys carefully
  • Use multiple connections for data loading
  • Include shard key predicates in queries
  • Consider reference tables for read-heavy, low-change tables
  • Balance connections across routers

The article demonstrates that while pgbench is optimized for single-node databases, careful adaptation can help benchmark and leverage the scalability of Aurora Limitless.