The longer the time series, the more accurate the trend extrapolations are. The minimum recommended time series length for accurate trend extrapolation is 3 months of consecutive data. Ideally, SQL Governor runs all the time in production environment and gathers workload data for many years.
However, good results can be had with a full month of data if the following criteria are satisfied:
- The data covers the busiest business period
- The scope of the project is 1-3 years
- The workloads are not growing exceptionally fast
- The trend extrapolation is manually set
For even shorter data collection periods (one or two weeks, preferably covering the change of the month), lower threshold levels should be set along with higher SLAs for adequate results.