A Cross-Generational Evaluation of SQL Aggregation Efficiency on Oracle Exadata Engineered Systems
Chaitanya Kulkarni, Chandrashekhar Medicherla, Bharadwaj Vulugundam, Tejas Patel, Sandeep Shivam, Amit Kumar Padhy
Abstract
This paper presents a comparative performance analysis of SQL aggregate functions across three generations of Oracle Exadata Database Machine: X8M, X9M, and X11M. Utilizing generation-aligned Oracle Database configurations (19c, 23ai, and 26ai), we executed a suite of TPC-H-derived decision support queries covering foundational aggregates (COUNT, SUM, AVG) and advanced OLAP operations (GROUP BY, ROLLUP, CUBE). The test environment was designed to quantify the cumulative impact of platform evolution integrating architectural enhancements such as Smart Scan offloading, Persistent Memory (PMEM), and RDMA over RoCE with database engine optimizations on query execution efficiency. The results reveal substantial generational improvements in response time and CPU offload efficiency. X9M delivered higher throughput through optimized Smart Scan and PMEM integration, while X11M consistently achieved up to 41.7% lower query latency compared to X8M. These gains are attributed to the synergy between high-bandwidth RDMA interconnects and storage server I/O offload rates exceeding 96.4%, which enable near-linear scalability for complex aggregations. These findings provide actionable insights for database architects and performance engineers evaluating Exadata refresh strategies. By quantifying the return on investment (ROI) in analytical workload acceleration and server resource optimization, this study informs planning decisions for mission-critical systems requiring high-throughput, low-latency SQL aggregation.