# Performance
Oxiland does not yet publish a budgeted benchmark suite. Treat the following as
operational guidance, not contractual numbers.
## Practical defaults
- Prefer streaming parse and serialize APIs for large files instead of loading
entire documents as strings when the API offers a stream or path path.
- Consume SPARQL and `find` iterators incrementally; avoid collecting unbounded
results into lists.
- Persistent (Fjall) models keep a full in-memory Oxigraph working set for
query—plan RAM for the dataset size in addition to on-disk footprint.
- Apply SPARQL `LIMIT` / application budgets before exposing query to untrusted
or latency-sensitive callers.
- Parse and bulk-load iterators do not expose wall-clock cancellation; stop
consuming them or isolate the work at the process/thread boundary.
## Related guides
- [Streams and iterators](streams.md)
- [Persistence](persistence.md)
- [Rust production](rust-production.md)
- [Python production](python-production.md)
- [FAQ](faq.md)