Optimizing CMS Data Formats for Analysis

This project examines two ways to improve CMS MiniAOD creation and analysis performance: one on job configuration level; the other on class definition level. Measurements of time to read various objects, size on disk and resident memory size are presented as Output Module parameters and PAT object definitions are inspected. We found that increasing basket size could lessen MiniAOD size and time to read whereas some others were set near the optimal point. Modifying string containing attributes also shows a slight decrease in size and time to read objects in comparison with the original MiniAOD structure.