I'm BFQ, I'm dynamite (or almost ...)
summary
BFQ now runs at 400+ KIOPS on commodity Intel CPUs, integrates with blk-mq/mmc, and lands a stack of work (about 37 new commits, 16 merged, 418 sketched patches) to cut scheduler overhead and tighten latency guarantees. App-start responsiveness shows dramatic wins—examples include LibreOffice startup from ~45s down to ~1s under heavy background I/O—and eMMC throughput jumps from ~7 MB/s to >15 MB/s in specific pathological workloads; demos and Phoronix/MMtests benchmarks back the claims.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:01I/O Scheduler role
- 4:25Block layer flavors
- 5:00Recent improvements
- 7:51Benchmarking efforts
- 9:40Performance results
- 15:46Demo and testing
- 18:58Next steps
- 21:03Enterprise storage challenge
- 31:29Conclusion
related talks
41mA Solution to High Latencies Caused by I/O
49mBFQ I/O scheduler: more throughput, control and efficiency
27mThe Two New I/O Controllers and BFQ
21mNVMe Emulation Performance Optimization
19mSynQuacer use case update
42mBoF: Realtime Linux
16mElectrode: Accelerating Distributed Protocols with eBPF
25mShenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads