Fail-stutter: an exact placement example ======================================= This is constructed arithmetic, not a storage benchmark, a reproduction of a 2001 implementation, or a recommendation to alter an existing RAID layout. Run with Python 3 (standard library only): python3 model.py --check python3 -m unittest -v test_model.py python3 -O -m unittest -v test_model.py Inputs: constant useful-logical-payload rates for mirrored pairs, in decimal MB/s, and a freely divisible logical payload in MB. Replication traffic is not counted twice as useful data. Capacity, placement, coordination and shared-link constraints are omitted. No completion/failure guarantee is modeled. For rates r_i and payload D, equal shares take max_i(D/(N*r_i)). Proportional shares D*r_i/sum(r_i) all finish at D/sum(r_i). The second time reaches the ideal capacity lower bound because no schedule can write more than t*sum(r_i) useful MB in t seconds under these assumptions. The test suite also recomputes a separate source correction: FAST 2018 Table 3 external rows sum to 44 of 112 root-cause occurrences, not 44 of 101 reports. No independence assumption, confidence interval or fleet rate is inferred. Sources: https://research.cs.wisc.edu/wind/Publications/fail-stutter.pdf (section 3.2) https://www.usenix.org/system/files/conference/fast18/fast18-gunawi.pdf (Table 3) Source review and example: Hesela, 2026-11-20. Definition: https://hesela.dev/terms/fail-stutter/