Are you interested in knowing how long a utility will take to run with an estimate from your own system’s utility history? Now you can!
Unlock the power of Db2 utilities history for a competitive edge with machine learning estimation!
Db2 utilities history introduced in Db2 for z/OS 13 FL 501 and enhanced with object-level history in FL 504 gives you access to a picture of your Db2 maintenance never available until now. The true power, though, comes when you unlock this data and leverage it to its fullest potential with machine learning for powerful analytics. Run the most streamlined and efficient maintenance windows possible and turn those savings into a competitive advantage for your business day. The future starts by learning from your history!
Db2 Utility Estimator, a feature of Db2 Automation Expert, leverages the saved utility history to build models used by the estimator to determine how your utilities process against the objects in your database. The initial iteration works on table space REORGs, but this is just the beginning. The model is trained using multi-variant linear regression and other techniques helping to correlate the attributes and sizes of your table spaces into an estimate of the execution times of similar table space objects.
The models for estimator are trained and reside in Unix System Services for z/OS (USS) so your data remains on the mainframe making the training and estimating very fast indeed. A new job extracts the utilities history data from the catalog tables and presents it to the estimator processes to train the models. The extracted data is then processed into one or more models, depending on the quality of correlation between a subset of the data not used for training and the estimates that the models would generate for that data. Since we are trying to estimate varying sizes of table spaces, the training process needs at least 500 rows of table space REORG data from the utilities history, with best results when more rows and a wider variety of objects are provided. The training is evaluated by the R2 regression coefficient and will warn you if the history does not correlate well. Since the portion of the data used to calculate the R2 score is randomly selected, retraining on the same data may improve the quality of the model if the initially selected data happens to not be representative.
Once the model is trained, you can then ask Db2 Automation Expert, at the time you generate your job profiles, to get the time estimates for the REORGs of each of the table spaces in your object profile and/or for the whole process.
This will show how well the items in your profile will fit in the maintenance window to which it will be targeted. We will show the calculated estimates for elapsed, CPU, and zIIP times.