Improving Longhorn Performance With SPDK
summary
Longhorn rewrites its engine to run as a single SPDK process per node, replacing per-volume iSCSI/TGT paths with a custom SPDK BDev and NVMe over Fabrics for replica transport. Benchmarks show writes and IOPS substantially increase versus current Longhorn, bandwidth matches or exceeds raw disk (over 1 GB/s) and latency drops from about 100 microseconds to 25–30 microseconds. Migration tooling and control-plane changes are required to reach feature parity; a technology preview targets upcoming releases.
AI-written summary · about
chapters (9)
- 0:00Introduction to Longhorn
- 1:00Longhorn architecture
- 7:13Current challenges
- 8:43Introduction to SPDK
- 12:37SPDK logical volumes
- 17:10Redesigning the data plane
- 22:21Performance results
- 27:18Future SPDK enhancements
- 29:44Roadmap and Q&A
related talks
32mLonghorn: Intro, Deep Dive and Q&A
31mLonghorn: Intro, Deep Dive and Q&A
32mLonghorn: Intro, Deep Dive and Q&A
47mLonghorn: Intro, Deep Dive and Q+A
37mBuild a Kubernetes Based Cloud Native Storage Solution From Scratch
48mIntro to Longhorn: Open Source Cloud-Native Distributed Block Storage Built On and For K8s
10mOvercome Interconnect Bottlenecks in HPC Clusters
27mThe Storage Architecture of Intel's Data Management Platform (DMP)