Benchmarks / methodology
Reproduction starts with the caveats.
The result set below is intentionally explicit about what was measured, what was calculated, and what still needs confirmation before an independent reproduction.
Evidence status
User-reported local development results
user-reported development measurement; raw artifacts for the matched comparison are pending publication. Existing repository logs are useful adjacent artifacts, but they do not establish the exact matched comparison described here.
Test machine
HP Omen Windows laptop
CPU model, RAM capacity, SSD, Windows build, MongoDB version, and Docker environment were not recorded in the supplied result set. They should be captured for a reproducible report.
LioranDB workload
Configuration recorded in the observation.
| documents | 10,000,000 |
| collection | Single collection |
| document Shape | { username, age } |
| liorandb Indexes | Secondary and text indexes disabled |
| liorandb Batch | 512 |
| liorandb Workers | 1 worker thread as reported |
| liorandb Nodes | 1 internal node |
| liorandb Partitions | 1 partition |
| durability | WAL enabled; checkpointing enabled |
What is calculated
- Full-run LioranDB average: 10,000,000 ÷ 36 = 277,777.78 docs/sec.
- MongoDB average: 10,000,000 ÷ 400 = 25,000 docs/sec.
- Peak-rate ratio: 300,000 ÷ 25,000 = 12×.
- Read ratio: 35,000 ÷ 16,000 = 2.1875×.
- Image reduction: 1 − 45 ÷ 164 ≈ 72.6%.
What is not claimed
- No universal “12× faster” statement.
- No simultaneous sustained read/write rate.
- No fsync-per-write or acknowledgement guarantee from “WAL enabled”.
- No memory-efficiency comparison; MongoDB memory was not measured.
- No inference from local Windows results to managed cloud throughput.
Limitations
A specialized raw-ingestion test is useful, not universal.
Secondary and text indexes were disabled for LioranDB, which reduces write amplification. MongoDB indexing semantics and write concern were not recorded in the comparison. The read result does not specify query type, indexing, thread count, cache state, or durability configuration. Simple documents are not a proxy for arbitrary production workloads.