hesela.devFlashBlox
FlashBlox is a research SSD virtualization design that assigns channels or dies to tenants and periodically migrates data to balance wear.
Channel isolation reserves transfer paths; die isolation still shares a channel bus. Soft-planes are software regions striped across physical planes, not independent NAND planes. [1]
Wear balancing swaps heavily worn resources with low-wear-rate ones. Requests to the erase-blocks being swapped wait; other blocks remain serviceable. The API requires log compaction aligned with the virtual SSD's erase granularity, and tenants cannot borrow one another's pre-erased blocks. [1]
Evidence boundary: the 2017 prototype's migration measurements are distinct from its simulated lifetime results. The reported 95% of ideal lifetime is a model result, not observed fleet survival or a guarantee against arbitrary changing adversarial traffic. [1]
Relations
- related
- Open-channel SSD, NAND channel, NAND plane, Wear leveling, Inter-flow interference, Slowdown fairness
Limitations
- Physical partitioning does not remove migration pauses or every device-wide shared resource. [1]
Citations
Further reading
Sources & caveats
- Full primary-paper review, including the adversarial-workload caveat in section 3.1.4 footnote 2. No FlashBlox execution, performance replication or device-lifetime measurement.