Performance and memory baselines (0.9)¶
Status: established for release decisions Related: VERIFICATION.md
Budgets below are tip-measured on Apple Silicon / Linux CI class machines for representative workloads. Regressions above the budget require a written decision in the milestone report (no silent waivers).
Workloads¶
| ID | Workload | Metric | Budget |
|---|---|---|---|
| P-C-01 | C world→parse Turtle (1k triples)→ASK | wall ms | ≤ 50 ms median |
| P-C-02 | C SELECT over 1k triples | wall ms | ≤ 80 ms median |
| P-S-01 | Fjall open + sync 1k quads | wall ms | ≤ 200 ms median |
| P-S-02 | redb open + sync 1k quads | wall ms | ≤ 250 ms median |
| P-M-01 | CAPI RSS after parse 10k triples | RSS MiB | ≤ 128 MiB |
Measurement method¶
- Warm once, then 5 timed iterations; report median.
- Use
cargo build -p oxiland-capi --releaseand defaultstorage-fjall(plus--features storage-redbfor P-S-02). - Record host, commit SHA, and feature set in the 0.9 compatibility report.
Overrun decisions¶
None open at 0.9 close. Future overruns must cite a budget ID and either fix, raise the budget with rationale, or accept a documented deviation.