Load testing Oracle databases
Originally published by TriCore: Sept 7, 2017
Oracle® Smart View provides a common Microsoft® Office interface designed specifically for Oracle's Enterprise Performance Management (EPM) and Business Intelligence (BI). By using Smart View, you can view, import, and create reports.
This blog discusses Oracle® load testing, also known as performance testing, which is used to test a database under various representative load conditions.
Introduction
Sometimes functional and regression testing is not sufficient for validating new code changes, especially when you are trying to make bulk code changes, doing a database or application upgrade, or changing any hardware. It makes sense to simulate the production load on the test environment to see how it behaves under various load conditions.
You might also perform a load test when a particular product is going live, or a new country localization is being implemented, in the case of an e-business suite.
There are four main phases in a load testing exercise:
1) Preparation
2) Simulation
3) Capture metrics
4) Analysis
The following sections explore each phase in detail.
Preparation phase
First identify the tool or procedure for conducting the test. Available tools, both open source and licensed, include the following software:
- Mercury LoadRunner - requires license
- Borland SilkPerformer - requires license
- HammerDB - open source
- Swingbench - open source
- Oracle Application Testing Suite (OATS) - requires license
You can also simulate load with custom-developed scripts.
Collect information
Collect information about the production system and the load test system. You need to know the operating system (OS) version, patch level, and hardware resources, such as central processing units (CPUs), random access memory (RAM), data storage device details, and so on.
Determine the metrics to be measured
List the metrics to be measured during the load testing exercise. Optional metrics include the following:
- CPU usage
- Load average
- Memory usage
- SQL response time
- Throughput
- Hits per second
You can configure OS watchers on the load testing system to get granular details of the OS and the hardware metrics.
The Oracle Automatic Workload Repository (AWR) needs to be configured, if it is not in place already.
Determine the core test areas
The core test areas for load testing depend on the type of application being hosted by the database:
- For an E-Business Suite (EBS) database, the month-end period is a major event that you can include during simulation load.
- For an EBS environment with manufacturing, the Advanced Supply Chain Planning (ASCP) plan run is a major event, which you need to test.
- You can simulate the payroll run of an application if it is being used in an Human Resources (HR) module.
- In the case of a core banking database, you can simulate the End of Day (EOD) processing to account for most of the load.
All of the preceding cases are more like batch processing scenarios. Along with these cases, you should also include transaction processing windows when more concurrent users are on the system.
Create a test plan:
Use the following stitems to create a test plan:
- Include at least two different phases for the test, a baseline and a post-upgrade phase.
- The baseline test is used to measure the current environment and validate it against the actual production environment. The post-upgrade test evaluates the environment after the upgrade is made.
- By using the baseline and the post-upgrade tests, you can analyze the impact of the upgrade.
- Identify load test procedures, including how to start the load test and how to evaluate test progress for each test.
- List the tests to be conducted.
- Plan the number of test run iterations for each test. For example, run the same test three times to get a more accurate average measure of transaction response time.
- Identify the reports for each of the load test phases. The reports should include information on results from each test run, metrics, and other collected information. Create report templates.
- Make a checklist of test actions that you need to execute after load testing. These can include steps to move the database to the pre-load testing state.
Some considerations
You need to consider the following issues if the load test system is not a 1:1 scale of the production system:
- Behavior in load testing is different than in production.
- Errors shown in production don't reproduce in load testing.
- Transaction performance is different between load testing and production.
- Other factors, not listed above, can also influence the load testing, and therefore, the results and analysis.
Simulation phase
After you finalize the test plan and choose the tools, you need to install and set up the necessary tools for simulation.
You can simulate load in different ways based on the tools being used. It is as simple as running a few select data definition language/data manipulation language (DDL/DML) queries on a set of tables or as complex as replicating a series of batch processing programs like ASCP run, Payroll run, or EOD processing.
First, capture and replay the baseline load from the source system in a load test. Then, move on to replaying the representative load.
HammerDB and OATS/Oracle Database Replay
HammerDB and OATS/Oracle Database Replay have the following characteristics:
- HammerDB enables you to capture the load on the source system by using Oracle trace files. Then, they can be moved to the load test system and replayed on it.
- Learn more about the detailed process.
- OATS is a comprehensive, integrated testing solution that ensures the quality, scalability, and availability of your web applications, web services, packaged Oracle applications, and Oracle databases.
- Oracle Database Reply is another option for performing load testing. You can use it from DBMS_WORKLOAD_CAPTURE APIs or Enterprise Manager. However, there are some licensing implications to use these tools. Using API’s, the DBMS_WORKLOAD_CAPTURE package is part of the Oracle Real Application Testing option. DBMS_WORKLOAD_REPLAY package is part of the Oracle Real Application Testing option. Use of the function DBMS_WORKLOAD_REPLAY.COMPARE_PERIOD_REPORT() also requires a license of Oracle Diagnostics Pack.
- You can also access the SQL Performance Analyzer feature from the SQLPLUS interface.
The DBMS_SQLPA package is a part of Oracle Real Application Testing option. The Oracle Real Application Testing license is required for both capture and replay, and systems for database replay is charged by the total number of CPUs on those systems. Refer the following My Oracle Support (MOS) note for detailed steps for capture and replay using API’s:
Using Enterprise Manager
You can perform capture and replay from Oracle Enterprise Manager (OEM) interface. From the Enterprise menu of the Enterprise Manager Cloud Control console, select Quality Management->Database Replay.
A Database Replay page appears, similar to the following image:
Learn about the detailed procedures for Capture from OEM and for Replay from OEM.
Capture metrics phase
While the simulation phase is running, make sure that the OS watcher is running and gathering statistics on the load testing environment.
Ensure that AWR snapshots are running at 30-minute intervals.
If an OEM infrastructure is available, ensure the OEM agent is active on the load testing server and that data is being captured.
Analysis phase:
To analyze the load test results, consider the following items:
Compare the baseline load and the representative load. Depending on the options available, you can analyze AWR reports and compare OEM data for various load situations.
Look for the peaks in CPU and memory usage, SQL response time, and throughput. Confirm that the peaks are within the permissible limits.
If you used Oracle Database Replay for simulating load, use this document to analyze the workloads.
If you used OATS for capture and reply, use this document to analyze the test results.
If you used the HammerDB tool, use this document.
Based on your analysis of the metrics, you can decide whether to go live with your planned changes or to adjust your plan.
Conclusion
The preparation phase is pivotal to the whole load testing exercise. Exercise more caution in the analysis and conclusion, especially when the load testing system is not a 1:1 replica of production.
You need to conduct each of the steps and processes carefully because any missed step or issue can invalidate the test results. This can result in a decision made under false or misleading information.
A successful and well-conducted load test exercise increases confidence and reduces the risk of a system upgrade.
|
|
|
|
|
|
If you have any questions on the topic, comment in the field below.