Skip to content

BENCHMARKS.md

Latest commit

 

History

History
55 lines (42 loc) · 1.7 KB

BENCHMARKS.md

File metadata and controls

55 lines (42 loc) · 1.7 KB

Benchmark Results

Measured on Apple M4 Max, Rust 1.85 stable, --release profile.

Run benchmarks locally:

cargo bench --all-features

Snapshot & Restore

Operation 100 orders 1,000 orders 10,000 orders
create_snapshot 127 µs 147 µs 329 µs
restore_from_snapshot 189 µs 308 µs 1.43 ms
enriched_snapshot (ALL) 129 µs 148 µs 334 µs
enriched_snapshot (MID_PRICE) 128 µs 148 µs 333 µs
snapshot_json_roundtrip 202 µs 1.76 ms —

Journal & Replay

Operation 100 events 1,000 events 10,000 events
journal_append 3.7 µs 22.6 µs 562 µs
replay_from_journal 84 µs 722 µs 7.88 ms
replay_verify 219 µs 889 µs —

Order Operations (existing)

Operation Time
Add limit order (single) ~2.1 µs
Match market order (deep book) ~34.6 ns
Mixed operations (1,000) ~172 µs

Serialization

Format Serialize Trade Deserialize Trade
JSON ~80 µs ~84 µs
Bincode ~89 µs ~98 µs

Observations

  • Snapshot creation scales sub-linearly: 10,000 orders only ~2.6x slower than 100 orders due to efficient SkipMap iteration.
  • Enriched snapshots add negligible overhead: metric calculation is fast compared to snapshot creation itself.
  • Journal append is very fast: ~3.7 µs for 100 events (37 ns/event) thanks to the in-memory journal implementation.
  • Replay throughput: ~1.27M events/sec at 10,000 events, suitable for fast disaster recovery.
  • Matching is sub-microsecond: market order matching on a deep book takes ~35 ns — well within HFT latency requirements.