FusionRAID: Achieving Consistent Low Latency for Commodity SSD Arrays
summary
FusionRAID is a RAID redesign that implements two-phase writes (temporary replicated writes then in-place parity conversion), a dynamic block map with static stripe mapping, and a lightweight sliding-window spike detector to redirect IO off slow drives. Evaluated on a 30-Intel-SSD array with traces and apps, it reports 22%–98% median latency reduction and 2.7x–62x (up to 35x reported) tail reduction, with temporary space overhead falling from >2× to 1.17× after automated stripe-ready replica placement and conversion.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:35All-Flash Arrays (AFAs) On Rise
- 1:24Severe SSD RAID Performance Problems
- 2:18Observations from Empirical Study
- 5:42Shared Storage Pool
- 6:41FusionRAID Optimized Writes
- 8:00Spike Detection and Request Redirection
- 9:03Evaluation Overview
- 9:38Evaluation: Trace-driven Workloads
related talks
53mOvercome Blast Radius Anxiety With Ultrafast Rebuilds and Performance for High-capacity SSDs
45mBig Data and SSDs
30mHybrid XFS: Idea to Exabytes in 12 Months
52mHow ZNS SSD's Are Changing the Landscape
20mMaximizing the Benefit of Flash Storage for Aerospike
18mAlleviating Garbage Collection Interference Through Spatial Separation in All Flash Arrays
29mLarge-Scale Benchmark Results of NVMe Over Fabric Technology
18mOPTR: Order-Preserving Translation and Recovery Design for SSDs with a Standard Block Device Interface