Tuning Principle

Performance tuning maximizes system efficiency to handle present and future needs by effectively configuring hardware, OS, and application settings.

To achieve the expected tuning outcome, you need to follow specific principles. The following are key tuning principles:

• Comprehensive analysis of resource bottlenecks: When analyzing system performance, use performance analysis tools such as nmon to examine the system resource utilization from multiple dimensions, because a performance dip in one part may result from bottlenecks in other parts or limited resources.
• Isolated parameter tuning: During performance parameter tuning, modify only one parameter at a time. For example, during Spark tuning, the parameters involved in HiBench include the shard count, memory capacity, and core count. Only one parameter can be adjusted at a time. This prevents confounding effects, ensuring that performance shifts can be accurately attributed to a specific change.
• Accounting for tooling overhead: Profiling tools consume system resources (CPU/Memory) themselves. Always account for this overhead, as the monitoring process might inadvertently exacerbate existing bottlenecks.