To simulate Oracle database workloads, we used SLOB version 2.4, which is a simple Oracle database schema generation and I/O testing toolkit. We used the SLOB multiple schema model to generate a 1.2 TB dataset (128 schemas x 9,600 MB each). We simulated an OLTP workload by configuring SLOB to generate a distribution of 80 percent reads and 20 percent write I/Os. For details on the SLOB parameter settings, see SLOB parameter settings (slob.conf).
The default schema that is generated by the SLOB tool contains a lot of redundant and repetitive data and, therefore, is highly compressible. To get more realistic results, especially for the deduplication and compression study, we modified the original SLOB dataset and replaced it with a unique dataset using a customized PL/SQL script that uses Oracle’s dbms_random.string function. For details on the custom script, see SLOB dataset customization.
We collected the database performance data through the Oracle AWR report. During each test, we configured AWR to capture data every 3 minutes. From the report, we extracted read, write, and total IOPS; average I/O latency (in ms); and CPU utilization percentage metrics.