Kernel Integration Benefits

OxiBase builds on research from CumulusDB(Leis and Dietrich), a unikernel-based DBMS that integrates kernel primitives for optimal performance, inspired by related work on OS-DBMS co-design(Giceva et al.), virtual memory snapshots(Kemper and Neumann), buffer management(Leis et al.), and OS-oriented systems(Li et al.)(Skiadopoulos et al.). Kernel integration provides privileged access to hardware and OS resources, enabling features that traditional layered architectures cannot achieve.

Privileged Hardware Access

• Direct NVMe Queue Access: Bypass OS storage layers for kernel-bypass I/O, reducing latency in high-performance SSD scenarios.
• Lock-Free Page-Table Walks: Concurrent, lock-free access to MMU structures for fast virtual-page presence checks (demonstrated 40x speedup in CumulusDB benchmarks(Leis and Dietrich)).
• IRQ and CPU Management: Manipulate interrupt vectors, block IRQs, and control CPU shutdown for query-plan-aware scheduling.

Zero-Copy Data Paths

• Stream data from query execution to network results without copying, leveraging hypervisor-shared memory regions(Leis and Dietrich).
• Elastic resource allocation through hypercalls for dynamic VM scaling(Leis and Dietrich).

Active Virtual Memory Management

• Virtual memory as an active abstraction, with inconsistent TLB states handled efficiently for concurrent OLTP/OLAP workloads(Leis and Dietrich).
• Advanced snapshots using ad-hoc parallelization and reader-side TLB invalidation, avoiding TLB shootdowns(Sharma et al.), with evaluation of virtual memory primitives(Loeck and Dietrich) and support for heterogeneous hardware(Mühlig et al.).

These benefits align with OxiBase’s unikernel compilation goals, enabling full hardware exploitation without OS overhead. See the roadmap for more details.

---

1. Leis, Viktor, and Christian Dietrich. “Cloud-Native Database Systems and Unikernels: Reimagining OS Abstractions for Modern Hardware.” Proceedings of the 50th International Conference on Very Large Data Bases [Guangzhou, China], 2024, pp. 2115–22, https://doi.org/10.14778/3659437.3659462.
2. Giceva, Jana, et al. “COD: Database / Operating System Co-Design.” CIDR, 2013.
3. Kemper, Alfons, and Thomas Neumann. “HyPer: A Hybrid OLTP&OLAP Main Memory Database System Based on Virtual Memory Snapshots.” ICDE, 2011, https://doi.org/10.1109/ICDE.2011.5767867.
4. Leis, Viktor, et al. “Virtual-Memory Assisted Buffer Management.” Proceedings of the ACM SIGMOD/PODS International Conference on Management of Data [Seattle, WA, USA], 2023, https://doi.org/10.1145/3588687.
5. Li, Qian, et al. “A Progress Report on DBOS: A Database-Oriented Operating System.” CIDR, 2022.
6. Skiadopoulos, Athinagoras, et al. “DBOS: A DBMS-Oriented Operating System.” PVLDB, vol. 15, no. 1, 2021, https://doi.org/10.14778/3485450.3485454.
7. Sharma, Ankur, et al. “Accelerating Analytical Processing in MVCC Using Fine-Granular High-Frequency Virtual Snapshotting.” SIGMOD, 2018, https://doi.org/10.1145/3183713.3196904.
8. Loeck, Yannick, and Christian Dietrich. “Evaluation and Refinement of an Explicit Virtual-Memory Primitive.” IEEE Access, vol. 11, December 2023, pp. 136855–68, https://doi.org/10.1109/ACCESS.2023.3338149.
9. Mühlig, Jan, et al. “Mxkernel: A Novel System Software Stack for Data Processing on Modern Hardware.” Datenbank-Spektrum, vol. 20, no. 3, 2020, https://doi.org/10.1007/s13222-020-00357-5.