modern-fs-benchmark collects results for multi-device CoW filesystems, filesystems that copy data on write, under workloads that classic benchmarks skip. The latest run is dated September 18, 2026, with 593 runs recorded and 145 trend points shown.

CI tests use loop devices, which are virtual block devices, on ephemeral VMs, with one VM per filesystem. The author recommends comparing result shapes and ratios rather than absolute MB/s.

In this index, a value of 100 corresponds to the median of the selected cohort, and the calculation uses eight recent complete runs from that cohort. Metrics are normalized by direction and combined as a geometric mean with equal subgroup and group weight, while integrity is never averaged.

Claim check:

  • modern-fs-benchmark collects results for multi-device CoW filesystems under workloads that classic benchmarks skip. (confirmed by the publication itself: evidence; «Multi-device CoW filesystems under workloads classic benchmarks skip»)
  • The latest run is dated September 18, 2026, with 593 runs recorded and 145 trend points shown. (confirmed by the publication itself: evidence; «latest run 2026-09-18 18:50:45 UTC, kernel 7.0.0-1012-azure, 593 runs recorded · 145 trend points shown»)
  • CI tests use loop devices on ephemeral VMs, one VM per filesystem. (confirmed by the publication itself: evidence; «CI runs use loop devices on shared ephemeral VMs (one VM per filesystem)»)
  • The author recommends comparing result shapes and ratios rather than absolute MB/s. (confirmed by the publication itself: evidence; «compare shapes and ratios, not absolute MB/s»)
  • In this index, a value of 100 corresponds to the median of the selected cohort, and the calculation uses eight recent complete runs from that cohort. (confirmed by the publication itself: evidence; «Score model v1: 100 = the selected cohort median. Metrics are normalized by direction, then geometric-meaned with equal subgroup and group weight. Using 8 recent complete selected-cohort runs»)
  • Metrics are normalized by direction and combined as a geometric mean with equal subgroup and group weight, while integrity is never averaged. (confirmed by the publication itself: evidence; «Metrics are normalized by direction, then geometric-meaned with equal subgroup and group weight. Using 8 recent complete selected-cohort runs; integrity is never averaged.»)

Primary sources:

score 64.6 out of 100 · kind: research