FLIN
FLIN is a research SSD-controller scheduler that reorders flash transactions to reduce interference between I/O flows while respecting host-assigned priorities.
Its three stages handle queue insertion, priority arbitration and read/write wait balancing, including garbage-collection work. Scheduling uses estimated transaction-alone waiting time, not a fresh isolated hardware run for each decision. [1]
Evidence boundary: the four-device interference experiment and MQSim evaluation of the proposed policy are separate. FLIN is not a standard NVMe feature or a host switch for arbitrary SSDs. [1]
Hesela's constructed baseline-sensitivity example holds shared latencies fixed while changing the reference means. Its fairness ranking reverses without a device becoming faster; this is arithmetic, not a FLIN replication or a measured estimator error. [2]
Relations
Limitations
- Relative fairness does not guarantee an absolute latency target. [1]
Citations
- Tavakkol et al. FLIN: Enabling Fairness and Enhancing Performance in Modern NVMe Solid State Drives. ISCA 2018, sections 2.2, 4-7; DOI: 10.1109/ISCA.2018.00041.
- Hesela: A changed baseline can reverse a fairness ranking
Sources & caveats
- Primary-paper review and exact constructed arithmetic; no SSD benchmark, firmware execution or scheduler replication.