JFS2/DIO Sequential Input/Output Performance on IBM pSeries 690
An IBM Redpaper publication
Note: This is publication is now archived. For reference only.
In this Redpaper we discuss the capabilities of the newest IBM pSeries 690 processor and its I/O subsystem. In the main section, we review the JFS2 (Enhanced Journaled File System) with Direct I/O implementation and provide details about our disk subsystem and disk adapters selections. We also provide guidance on selecting the proper RAID configuration, and give other performance and tuning tips.
In Appendix A we explain array controller concepts; in Appendix B we address FAStT setup concepts; and in Appendix C we provide a brief overview of some of the available technologies and products that we considered for our benchmark, and explain why we chose a FAStT600 Turbo configuration for this sequential I/O workload benchmark.
Note that because each environment is unique, and because technologies and capabilities are continuously changing, this discussion should only be used as a broad guideline of considerations to keep in mind when choosing storage.
The goal of this high performance I/O benchmark was to demonstrate that the IBM pSeries 690 has a very powerful I/O subsystem for handling sequential I/O workloads. The study shows that an appropriately configured pSeries 690 system, combined with a properly sized disk subsystem, is capable of sustaining 18.5 gigabytes per second (GB/s) sequential I/O when utilizing disk controller cache. When performing disk I/O, the server can sustain 16 GB/s aggregate sequential read or 14.8 GB/s aggregate write performance using either striped logical volumes or Enhanced Journaled File System (JFS2) using Direct I/O.