Cumulus: Co-Designing Unikernels and Database Systems for the Cloud

• Applicants: Professor Dr. Viktor Leis; Professor Dr.-Ing. Christian Mengert-Dietrich
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems ; Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 562906980

More and more organizations are moving their IT infrastructure to public clouds. In this context, database systems are of particular importance, as they are a foundational component for higher-level applications. The worldwide database management system (DBMS) market revenue for 2022 is estimated at $91 billion, 55 percent of which is spent in the cloud. This implies that database systems are frequently a performance bottleneck and cost factor, which makes it crucial that they can fully exploit the expensive cloud hardware they run on. The Cumulus project focuses on cloud-specialized database systems that execute in virtualized cloud environments and that make full use of directly-attached hardware resources. We explore DBMS-OS co-design principles for a deep mutual integration of DBMS and operating system, a method we term "kernel integration", that establishes tight control cycles between task execution, storage and network access, and memory allocation. To facilitate interaction, we take inspiration from and fully utilize modern asynchronous hardware interfaces and their high degree of parallelism. The project will contribute to the long-standing debate regarding whether a DBMS should utilize existing OS mechanisms (e.g., for thread scheduling, caching) or implement its own OS-like services (e.g., user-space task queues, buffer manager). For performance reasons, database architects are currently often forced into the latter option, having to implement tasks such as scheduling and hardware management that traditionally were performed by the OS. We aim to demonstrate that DBMS/OS co-design can overcome this situation and make high-performance database systems both faster and simpler.

• DFG Programme: Research Grants