Workload
The workload used in this study emulated a business processing transactions generated by many active users. These users accessed a single database to process warehouse orders, manage inventory, and bill customers.
Workload description
The workload was configured with a zero amount of think time, the pauses between events that occurs when humans are interactively starting transactions. This results in an intensive load from a single user, so that with the 20 - 100 users used in the test environment the workload processed is similar to a much larger number of connected users in a real setting. The level of utilization could be compared to that of a database working as the backend of an application server, or a transaction concentrator.
Workload characteristics
The workload used in this study emulated a small number of warehouses, to intentionally create cluster contention on the warehouse table. This forced a high level of cluster internal communication used to ensure cache consistency and lock handling. This allowed good monitoring of the impact of the various setup modifications.
Performance tools used to analyze the workload
The study used the Oracle AWR reports as a starting point for performance analysis. If contention or excessive latencies are taking place, they show up in the Top 5 wait events as a high percentage of the run time, and they can be analyzed further with the detailed information such as the relevant wait events or the cache information.
SADC and SAR were the primary tools used to evaluate the Linux™ performance.
The netstat -s command was used for reports of I/O re-sends, to ensure that the TCP/IP communication is not limited by the interface capacity.