Benchmarks

1M documents (employee-shaped, ~500B each). Docker-Compose, in-network runner (container-to-container — no port-forward artifact). AMD EPYC-Genoa, 4 vCPU, 8 GiB RAM. Source: tests/comparison-mongodb/.

OxiDB vs MongoDB 7 (1M docs)

OxiDB 24 — MongoDB 0

Across the full 24-test suite at 1M documents, OxiDB wins every measured workload. Largest wins are on counts (index-only path) and aggregation; smallest wins are insert (1.1×) and bulk full scans (1.1–1.3×).

Operation	OxiDB	MongoDB	Winner
Count all (1M docs)	200µs	429ms	OxiDB 2189×
Top 5 cities (aggregation)	1ms	822ms	OxiDB 1262×
Count age ≥ 50	200µs	156ms	OxiDB 645×
Count dept=Engineering	300µs	53ms	OxiDB 197×
Compound (composite idx) dept+status	161ms	626ms	OxiDB 4.1×
Random seq lookup × 100K	8.07s	27.6s	OxiDB 3.4×
Indexed: salary 80K-120K (235K results)	1.01s	3.25s	OxiDB 3.2×
Indexed: city=Tokyo (100K results)	448ms	1.37s	OxiDB 3.0×
Indexed: age ≥ 60 (300K results)	1.43s	4.12s	OxiDB 2.9×
Indexed: dept=Engineering (199K results)	1.17s	2.65s	OxiDB 2.3×
Create 4 indexes on 1M docs	2.82s	7.33s	OxiDB 2.6×
Bulk delete (10% of corpus)	5.88s	13.3s	OxiDB 2.3×
$or city=Tokyo OR Paris (no idx)	1.48s	3.15s	OxiDB 2.1×
Compound full scan (no idx)	515ms	1.03s	OxiDB 2.0×
$in country (no idx)	2.21s	4.03s	OxiDB 1.8×
Range (salary 50K-100K) no idx	2.06s	3.95s	OxiDB 1.9×
Exact match dept=Engineering (no idx)	2.18s	2.95s	OxiDB 1.4×
Match + Group (aggregation)	247ms	336ms	OxiDB 1.4×
Nested address.zip range (dot-path)	1.47s	1.91s	OxiDB 1.3×
Range query × 10K (10.2M rows total)	30.2s	35.9s	OxiDB 1.2×
Bulk insert 1M docs (batch 1000)	16.0s	18.3s	OxiDB 1.1×

Resource footprint (post-bench, 1M docs)

Resource	OxiDB	MongoDB	Note
Memory (RSS)	1.71 GiB	1.00 GiB	OxiDB caches are env-tunable (`OXIDB_DOC_CACHE_SIZE`, `OXIDB_DOC_BYTES_CACHE_SIZE`); MongoDB WT cache capped at 0.5 GiB.
Disk (data dir)	741 MB	626 MB	OxiDB stores JSONB; MongoDB stores compressed WiredTiger. ~18% larger on disk.

Methodology

Apples-to-apples networking. The Go bench runner runs as a container on the same Docker network as the engines (BENCH_MODE=innetwork). Host-mode runs through Docker's port-forward add ~160µs/round-trip that distorts small-payload workloads.
Single cold run. Each test runs once on a freshly-inserted 1M-doc corpus. No warmup.
Default configurations. No tuning. OxiDB uses defaults; MongoDB uses WiredTiger with --wiredTigerCacheSizeGB 0.5 per the existing compose config.
Same seed. Both engines see the identical document corpus (deterministic generator, fixed RNG seed).

OxiDB vs PostgreSQL (100K docs)

OxiDB 10 — PostgreSQL 10

Tied overall at 100K docs. OxiDB wins on document queries, counts, and aggregation. PostgreSQL wins on bulk insert, indexed range queries, and JOIN-heavy workloads. (Pending re-run at 1M scale.)