Baseline Vs. Guessing
The Accidental DBA, Monday, March 2nd, 2026
Too often 'it's slow' is a guess, not a diagnosis. Without historical performance data, you have no baseline to compare against. A single moment in time doesn't tell you whether your current state is normal, trending worse, or an outlier event.
4 Metrics Every SQL Server Should Track
To build a reliable baseline, make sure you're capturing at least these core counters regularly:
- CPU Utilization - High CPU could be steady load, runaway queries, or underlying hardware limits.
- Wait Statistics - Know what resource waits are accruing. They tell you what subsystem is stressed.
- I/O Latency (Reads/Writes) - Slow storage is a common bottleneck. Historical latency profiles help you spot regressions.
- Memory Metrics - Track buffer cache hit ratios, memory grants, and paging. Changes here often signal plan or workload shifts. Page Life Expectancy does not have a 'best practice' as it is a measurement, not a setting. 300 seconds is marketing nonsense. See what yours is during normal traffic over multiple measurements, and THAT is your baseline to compare against.
These metrics give you a multidimensional view of performance instead of a snapshot that might mislead.