Benchmark Results
All benchmarks run against a debug build (cargo run --bin benchmark_compare) on a local macOS machine. RookDB is compared to a local PostgreSQL instance.
Setup
| Parameter | Value |
|---|---|
| Rows | 100 000 |
| Iterations | 5 |
| Update predicate | id <= 50000 |
| Update assignment | score = score + 10 |
| Delete predicate | id <= 20000 |
| Build profile | debug |
| PostgreSQL schema | public |
The RookDB seed file (100 000 rows) is copied fresh before each iteration so every sample starts from an identical table state.
RookDB Summary (debug build, rook-only)
| Operation | Average | Fastest | Slowest | Total |
|---|---|---|---|---|
| UPDATE | 3 310.813 ms | 3 128.560 ms | 3 463.800 ms | 16 554.066 ms |
| DELETE | 98.649 ms | 97.017 ms | 101.215 ms | 493.244 ms |
| COMPACTION | 17.747 ms | 13.284 ms | 31.394 ms | 88.737 ms |
PostgreSQL Baseline (same conditions)
| Operation | Average | Fastest | Slowest | Total |
|---|---|---|---|---|
| UPDATE | 201.111 ms | 168.552 ms | 233.010 ms | 1 005.556 ms |
| DELETE | 5.224 ms | 4.902 ms | 6.248 ms | 26.120 ms |
| VACUUM (std) | 5.158 ms | 4.535 ms | 6.577 ms | 25.789 ms |
| VACUUM FULL | 40.812 ms | 39.361 ms | 42.185 ms | 204.061 ms |
RookDB vs PostgreSQL — Average Latency Comparison
| Operation | RookDB avg | PostgreSQL avg | Result |
|---|---|---|---|
| UPDATE | 3 310.813 ms | 201.111 ms | PostgreSQL ~16.5× faster |
| DELETE | 98.649 ms | 5.224 ms | PostgreSQL ~18.9× faster |
| COMPACTION vs std VACUUM | 17.747 ms | 5.158 ms | PostgreSQL ~3.4× faster |
| COMPACTION vs VACUUM FULL | 17.747 ms | 40.812 ms | RookDB ~2.3× faster |
Compaction Analysis
RookDB compaction is a full page rewrite — it visits every page that has dead tuples and rewrites it from scratch. This is architecturally equivalent to PostgreSQL's VACUUM FULL (which also rewrites the entire table and holds an exclusive lock). Against that baseline, RookDB is ~2.3× faster.
PostgreSQL's standard VACUUM is an incremental lazy sweep — it reclaims dead tuple slots in-place without rewriting the full page, which is why it completes in ~5 ms. RookDB does not yet have an equivalent incremental mode.
UPDATE Analysis
RookDB UPDATE is significantly slower because each matched row requires:
- A page-level write lock acquisition.
- A soft-delete of the old slot.
- An FSM-guided heap insertion for the new version (which may land on a different page).
PostgreSQL's UPDATE is an in-place MVCC version replacement that avoids an extra heap insert for rows that fit in the same page.
How to Reproduce
# RookDB-only benchmark (no PostgreSQL required)
cd code
cargo run --bin benchmark_compare -- --rook-only
# Full comparison against local PostgreSQL
cargo run --bin benchmark_compare -- \
--pg-url "postgresql://<user>@localhost:5432/postgres" \
--pg-schema "public"