Open-Source columnar file formats such as Apache Parquet and ORC have been widely adopted for data analytics systems. These formats, however, were developed more than a decade ago. Both the hardware and workload landscapes have changed since then. In this talk, I will first discuss some of the lessons learned from our comprehensive benchmark on Parquet and ORC. Then I will introduce F3, the next-generation data file format developed in our group, with an emphasis on its efficiency, interoperability, and extensibility.
