Reducing Relational Database Performance Bottlenecks Using 3d Xpoint Storage Technology

Relational database systems are data-intensive applications because of their heavy I/O workload during a query execution. With more and more data being generated, database sizes are increasing significantly. As a result, database systems incur a tremendous latency to return the query results back to the clients. Using the TPC-H benchmark program, this work explores several key issues that degrade database performance, including read amplification, inefficient use of temporary tables, and higher buffer pool cache miss rates. We show that when clients execute database queries on a conventional storage device, such as a solid-state disk (SSD), the extra I/O latency resulting from the above issues leads to bottlenecks that limit overall database performance. The availability of a new generation storage device, the Optane, which uses a new 3D cross-point technology, provides a feasible approach to accelerate the overall database performance. Processing database queries on this new device produces 1.5-8.3X improvement in the I/O response time, and 1.1-6.5X improvement in the total response time, compared to an SSD-based system.