hesela.devFreeblock scheduling
Freeblock scheduling fits background HDD transfers into predicted rotational delays while aiming to preserve the next foreground transfer's timing.
The foreground request is selected first; background work uses the remaining positioning opportunity. Extra seeks consume some of that opportunity. This is not simply waiting until the request queue is empty. [1]
Large sets of unordered blocks give the scheduler more choices. The last blocks of a scan can be awkward to reach; promoting them to foreground priority trades scan completion against foreground interference. [1]
The FAST 2002 external prototype used limited command queueing and adjacent-request merging to handle timing and drive-cache behavior. Imperfect prediction and control reduced usable bandwidth relative to the earlier simulation. [2]
Relations
Limitations
- Prediction errors can miss the intended sector passage and add a rotation of delay. [2]
Citations
- Lumb et al. Towards Higher Disk Head Utilization: Extracting "Free" Bandwidth from Busy Disk Drives. OSDI 2000, sections 2, 5.2 and 6.
- Lumb, Schindler and Ganger. Freeblock Scheduling Outside of Disk Firmware. FAST 2002, pp. 275-288; sections 2-5.
Further reading
Sources & caveats
- Historical source synthesis, not a current Linux scheduler option, SSD capability or reproduced benchmark.