StRAID: Stripe-threaded Architecture for Parity-based RAIDs with Ultra-fast SSDs
summary
StRAID implements a one-for-one stripe workers model that replaces MD’s centralized stripe list and heavy spinlock use, adds a parity cache and a two-phase stripe submission to aggregate partial-stripe writes, and uses a stripe state table for lockless coordination. A prototype is evaluated on NVMe SSDs, PM and RAM-disk with microbenchmarks and real traces, showing up to 5.8× write throughput vs Linux MD, up to 90% reduction in lock overhead, much lower CPU contention, and consistent read performance.
AI-written summary · about
chapters (14)
- 0:00Start
- 0:49RAID Systems
- 3:56SSD Storage Trend
- 4:35Linux MD upon SSDs
- 5:51Multi-thread Write Scalability
- 7:53Analysis of MD
- 14:57StRAID overview
- 18:59StRAID's Concurrency Control
- 23:05Evaluation Setup
- 24:11Micro-benchmark Results
- 25:28Breakdown of CPU cycles
- 26:59Upper bound Evaluation
- 28:43Macro-benchmark Results
- 29:09Conclusion
related talks
23mBtrfs RAID on zoned devices
24mMD cluster raid
48mSoftware Defined Storage the Linux way
29mDC-Store: Eliminating Noisy Neighbor Containers using Deterministic I/O Performance
24mLatest storage status on Arm
16mRearchitecting Buffered I/O in the Era of High-Bandwidth SSDs
17mHow advancements in storage are triggering innovations in open-source and vice-versa
21mSpool: Reliable Virtualized NVMe Storage Pool in Public Cloud Infrastructure