Disk-head time
Disk-head time is an estimate of HDD backend work combining per-fetch positioning cost with the transfer cost of bytes actually fetched.
The author explanation treats setup and transfer costs together; reducing request misses alone can spend more bandwidth. Demand hits and extra prefetched bytes must therefore be accounted for separately. [1]
Baleen reports the maximum backend utilization over ten-minute windows. Its admission and prefetching policies optimize average disk-head time, not an explicit peak-admission schedule. A pooled average cannot recover a maximum without the window history. [1]
A Hesela constructed ledger gives 90% demand hits with 150 ms of modeled work and 95% hits with 250 ms when expanded fetches transfer more bytes. This is exact hypothetical arithmetic, not a device measurement or a Baleen performance result. [2]