Describe the symptom precisely
Identify the affected function, time window, users and expected behavior. “The database is slow” is less useful than a repeatable report, screen or batch process with a measurable delay.
Separate workload from resource pressure
Review execution plans, wait events, I/O, CPU, locking and connection behavior in the context of the workload. A slow query, competing batch process and storage pressure require different responses.
Test one informed change at a time
A change should have a reason, a rollback path and a way to observe its effect. Multiple simultaneous changes make it difficult to know what improved or introduced a risk.
Practical example
Example: a slow month-end report
Compare the report’s execution plan and elapsed time with a normal period. Then check whether its data volume, statistics, joins or concurrent workload changed before selecting an intervention.
Best practices
Put the fundamentals in place.
- Capture a baseline before altering indexes, parameters or queries.
- Review execution plans using representative bind values and data volumes.
- Document the symptom, observation, change and post-change result for future support.
Frequently asked questions
Is adding an index always the answer to a slow query?
No. Indexes can help some access paths but may add write cost or be ignored by the optimizer. The plan and workload should guide the decision.
Why does workload timing matter?
A query that performs well alone may be affected by concurrent jobs, locking or resource contention during a specific business window.
Related RSAInfosys expertise