...
- Reference test cases provide an idea about the achievable performance for a given use case.
- They are not intended to guarantee absolute reference values as such values depend on the system environment and resources. Instead they provide information on the ratio of reference values when using different optimization settings.
- Consider the article Performance Optimizaton Optimization for API jobs when applying optimizations to your environment.
...
- Three test suites are provided:
- Test Suite 1: Zero Optimization
- Optimization: Jobs are configured without optimization, i.e. no tasks are pre-loaded or re-used
- Configuration:
<job tasks="1"/>
- Test Suite 2: Full Optimization
- Optimization: Jobs are fully optimized, i.e. tasks are pre-loaded and re-used for processing of parallel orders
- Configuration:
<job tasks="3" min_tasks="3"/>
- Test Suite 3: Balanced Optimization
- Optimization: Jobs are medium optimized, i.e. tasks are not pre-loaded but re-used
- Configuration:
<job tasks="3" idle_timeout="30"/>
- Test Suite 1: Zero Optimization
- Recommendation
- Try to use balanced optimization for your environment.
- There is only a small delay when initially loading tasks for the first order of a job chain.
- When running a medium or high number of orders then the performance is approx. equivalent to full optimization.
- This optimization respects use of resources (memory) by unloading tasks that are not used for a longer time (idle timeout).
- For high performance environments use the full optimization scenario.
- Guarantee that resources (memory) are available to pre-load and to retain the required number of tasks for parallel orders.
- Should available memory be exhausted then this scenario will create extremely poor performance due to memory swapping.
- Try to use balanced optimization for your environment.
...