| After loading and initial maintenance | SQLite | PostgreSQL |
|---|---|---|
| Messages | 4,360,000 | 4,360,000 |
| Database size, decimal GB | 8.008 | 14.324 |
| Database engine | 3.53.3 | 17.11 |
| Full-text implementation | FTS5 | tsvector with simple dictionary and GIN |
Results
All 11,640 measured operations completed without errors. The tables pool three rounds. Lower latency and higher throughput are better. FTS p95 is the maximum of only 15 observations per backend/query; see the sampling caveat below.
FTS5 versus PostgreSQL full-text search
| Workload | SQLite median / p95 (ms) | PostgreSQL median / p95 (ms) | SQLite ops/s | PostgreSQL ops/s |
|---|---|---|---|---|
Common: discussion (~25% match) |
19,220.321 / 21,491.441 | 22,335.698 / 24,793.862 | 0.050 | 0.044 |
Medium: deployment (~1.09%) |
1,292.016 / 1,334.494 | 201.237 / 212.263 | 0.896 | 4.9 |
Rare: quasar (~0.01%) |
6.045 / 6.504 | 22.672 / 23.708 | 162.7 | 44.3 |
Prefix: deplo (~1.09%) |
1,293.404 / 1,304.671 | 201.635 / 203.410 | 0.898 | 5.0 |
AND: deployment rollback (~0.099%) |
59.042 / 59.787 | 36.978 / 38.430 | 16.9 | 27.0 |
deployment in channel 042 |
127.693 / 130.478 | 35.095 / 427.356 | 7.8 | 10.9 |
Missing: absentlexemezzz |
0.314 / 0.488 | 22.065 / 23.419 | 3,076.1 | 45.4 |
quasar, offset 100 |
6.302 / 7.276 | 22.605 / 24.119 | 157.6 | 44.4 |
History and writes
| Workload | SQLite median / p95 (ms) | PostgreSQL median / p95 (ms) | SQLite ops/s | PostgreSQL ops/s |
|---|---|---|---|---|
| Latest 100 messages | 0.965 / 4.304 | 1.561 / 2.171 | 583.7 | 613.7 |
| 100 messages halfway through history | 20.268 / 36.208 | 65.703 / 128.343 | 48.7 | 13.1 |
| Latest 100, eight concurrent callers | 7.814 / 9.170 | 11.934 / 13.005 | 1,007.4 | 665.3 |
| Insert one message | 0.520 / 2.503 | 9.327 / 10.075 | 699.0 | 114.9 |
| Insert, eight concurrent callers | 4.612 / 44.283 | 74.908 / 87.154 | 712.5 | 109.2 |