Skip to main content Skip to main navigation

Publikation

An Evaluation of NVMe-over-Fabrics for Disaggregated Databases over Fast Networks

Jigao Luo; Nils Boeschen; Tobias Ziegler; Carsten Binnig
In: Meike Klettke; Ralf Schenkel; Andreas Henrich; Daniela Nicklas; Maximilian E. Schüle; Klaus Meyer-Wegener (Hrsg.). Datenbanksysteme für Business, Technologie und Web (BTW 2025), 21. Fachtagung des GI-Fachbereichs ,,Datenbanken und Informationssysteme" (DBIS), 03.-07, März 2025, Bamberg, Germany, Proceedings. GI-Fachtagungen (BTW), Pages 113-125, LNI, Vol. P-361, Gesellschaft für Informatik e.V. 2025.

Zusammenfassung

With increasingly faster networks, storage disaggregation in databases enhances resource utilization by independently scaling compute and storage nodes. Storage technologies like NVMe SSDs provide high throughput and low latency, but maintaining these characteristics in a disaggregated architecture poses significant challenges. NVMe-oF is a recent remote storage protocol that leverages fast network technologies such as RDMA for efficient storage disaggregation with low access overheads. In this paper, we present a first evaluation of the performance properties of NVMe-oF for database workloads to guide the development of databases. Our evaluation shows that NVMe-oF can offer high I/O performance when choosing optimal configurations for various database workloads. However, we also show that some optimizations, such as target-offload, which minimize CPU involvement on the storage side, are not (yet) working as expected.

Weitere Links