STOW
STOW is a write-cache policy that separates sequential and random write groups and adapts their destage order, concurrency and cache allocation.
Two address-sorted queues retain WOW-style reference bits. Occupancy thresholds control concurrent destages; hysteresis batches work from one queue to limit switching. [1]
The batch ends after a page-count threshold or sufficient growth in either queue. The count is pages, not time or write groups. [1]
The linked RAID analysis provides a switching-cost example with explicit assumptions. [2]
Relations
Limitations
- The rotating-disk evaluation used volatile DDR2 cache and an SPC-1 Like prototype workload, not a power-loss survival test. [1]
Citations
- Gill, Ko, Debnath and Belluomini. STOW. USENIX ATC 2009, sections 3-6 and Algorithms 1-3.
- Hesela: RAID write penalty is not a speed ratio, destage batching example.
Sources & caveats
- Primary-source review of the algorithm and evaluation. The linked arithmetic is a synthetic scheduling illustration, not a STOW implementation or replicated benchmark.