Customizing the System Stack for Data Processing on Modern Hardware

Modern data analytics and data science are at the heart of enterprise computing and drive advances in many scientific disciplines. In addition to designing new models and techniques for analyzing the data deluge, we also need system support that makes the analysis more performant and efficient when executed on modern and future hardware. Addressing such a challenge requires an effort that is beyond what can be typically done withing a single layer of the system stack: applications are typically unaware of the run-time state of a machine and the OS does not know what are the goals of the applications. In my talk, I will give an overview of my ongoing research on "Database/Operating system Co-Design" and show the benefits of a holistic approach by opening the interfaces and customizing the system stack for modern data processing workloads.