ZERO-COPY MERGE FOR ROOT RNTUPLE

Data sets in High-Energy Physics (HEP) are often produced in a parallel manner where tens or hundreds of processes produce partial results that are merged into a single file at the end of the data processing workflow. This is enabled by the parallel nature of HEP data, where individual collisions can be processed independently from each other. The predominant file format for storing HEP data is provided by the ROOT libraries. Modern file systems allow for zero-copy cloning of data blocks. This feature has a potentially high impact for the merging of HEP data, as it may allow creation of merged results without physically copying the partial results. The aim of this project is to evaluate the zero-copy merging functionality in modern file systems as well as the requirements to use it at the file format level (e.g., relocatability). The results of this project have an impact on ROOT’s object store support as well, as HEP data in object stores such as Intel DAOS is by nature prepared to exploit zero-copy merging