Enabling High-Performance Storage for RISC-V
summary
Lustre client ported to RISC-V by adding architecture support and three core changes (build target, 64-bit erasure coding paths, fallback header) and landing 7 upstream patches merged to master. Testing uses two QEMU RISC-V VMs (Ubuntu 24.04, kernel 6.8.4) plus an x86_64 server, module loading, mount, file IO and standard benchmarks pass, while MGS/MDS/OSS remain on other architectures. CI with Gerrit/CloudV catches regressions moving forward.
AI-written summary · about
chapters (15)
- 0:00Introduction and speaker intro
- 0:21Open-Silicon and company overview
- 1:02RISC-V readiness for HPC
- 2:24Lustre as HPC storage solution
- 3:53Lustre architecture components
- 5:00Identifying architecture gaps
- 6:43Patch separation and compatibility
- 8:08Core RISC-V patches for Lustre
- 9:33Dependency matrix and hardware constraints
- 10:15Testing with QEMU and CI/CD
- 11:17Test results and benchmarking
- 12:19Collaboration with Lustre community
- 13:25Upstream journey and timeline
- 13:47Future-proofing with CloudV
- 14:31Summary and closing remarks
related talks
18mRISC-V Support for Persistent Memory Systems
24mReliability, Availability, and Serviceability (RAS) on ARM64 status
50mOverview of the Linux Storage Stack
27mPreview: Storage
34mEnhancing Data Integrity in Linux
33mScaling KV Caches for LLMs: How LMCache + NIXL Handle Network and Storage...
26mLessons Learned: Optimizing KVM Performance for EHR Systems
27mrpdfs: Designing a High Performance File System
