Measure thrice, check twice, cut once

You can review the information on tuning when you have performance issues.

An old adage in carpentry is to Measure Thrice, Check Twice and Cut Once.

In the heat of a performance problem, it is very tempting to try different tuning parameters without fully understanding the root-cause of the problem. Some changes or combination of changes can have a negative impact on the system. After trying many tuning changes, it is possible that some of the non-beneficial changes are not rolled back.

From past experiences, we found that system optimization is often more effective and efficient if the problem is correctly analyzed and the root cause clearly identified. An approach that we have adopted is as follows:

• (measure) Measure the system to establish the baseline performance and throughput.
• (measure) Measure the system when performance issues arise.
• (check) Given the symptoms, formulate theories as to the root-cause of the problem and the potential tuning changes.
• (check) Ensure you can explain why certain tuning recommendations can help alleviate the problem and to formulate the expected behavior if the tuning change is applied.
• (cut) Make one (or a few) tuning change at a time - in some cases, multiple changes could negate the benefit of other changes.
• (measure) Measure the system and see if the system gained the intended benefits.