linnos · research

LinnOS

LinnOS is a research operating-system extension that classifies an incoming SSD read as fast or slow to guide admission and replica failover.

Related Observatory article: LinnOS on hesela.com

The OSDI 2020 prototype learns per workload-device pair from recent completions and pending I/O. A latency-critical read predicted slow is rejected before entering the device queue; the application can retry another replica. Its last-replica attempt bypasses rejection. [1]

The label is not a drive-failure probability or a latency guarantee. Live completions cannot reveal false revokes because rejected reads never reach that device; the paper evaluates those errors offline. LinnOS+HL additionally uses hedging and must not be confused with the raw classifier. [1]

Relations

related
Fail-slow, Tail latency, Hedged request, Adaptive replica selection

Limitations

  • Correlated delays, retraining needs and inference overhead limit transfer to other workloads and devices. [1]

Citations

  1. Hao et al. LinnOS: Predictability on Unpredictable Flash Storage with a Light Neural Network. OSDI 2020, pp. 173-190; sections 3-6. (accessed 2026-11-08)

JSON: /data/terms/linnos.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. LinnOS. Hesela. 2026-11-08. https://hesela.dev/terms/linnos/