Skip to content

BENCHMARKS.md

Latest commit

 

History

History
21 lines (16 loc) · 1.23 KB

BENCHMARKS.md

File metadata and controls

21 lines (16 loc) · 1.23 KB
Raw

Benchmark summary — alpha45.9 to alpha49

Scenario CLD2 warm zsync casync OSTree DVC OCI single OCI per-file
normal 419 239.827 10.986 196.227 67.109.107 196.511 196.733
high-entropy 67.108.937 67.342.559 67.160.433 67.130.107 67.109.076 67.130.606 67.130.785
small-files 2.147 3.515.724 55.116 34.554 6.874.124 188.165 73.044
heavy-change 26.845.855 27.078.879 26.871.717 26.971.645 67.109.040 26.971.886 26.972.139

Metric caveats

  • CLD2 warm: validated warm update bytes from alpha45.8 same-run consolidation.
  • zsync: HTTP bytes from the validated same-run baseline.
  • casync: HTTP .caidx/.castr extract with local seed directory.
  • OSTree: HTTP pull of v2 after v1 was already pulled; no static-delta generated in the first baseline.
  • DVC: new bytes added to a local DVC remote; this is not an HTTP byte benchmark.
  • OCI single/per-file: deterministic tar.gz layer model; not a real Docker daemon pull.

Conclusion

The results support a limited but useful claim: CLD2 is promising for cached update distribution with high reuse. The results do not support a universal superiority claim.