Benchmarks

Tested on 1.1M Wikipedia articles (1M English + 100K Japanese, avg. 666 chars per article) with MySQL 8.4 FULLTEXT (ngram parser). MygramDB v1.5.0 with verify_text: all and query cache disabled. All numbers are p50 over 10 iterations.

Reproducible

Run these benchmarks yourself: make bench-up && make bench-run in the repository.

Note on hardware

Measured on Apple M4 Max with 128GB unified memory. Unified memory has higher bandwidth than typical server DDR4/DDR5. On server hardware, absolute latencies will be higher for both engines, but the relative performance difference remains consistent.

Search Latency (SORT id LIMIT 100)

Search Latency (p50, log scale)

Query Type	Matches	MySQL	MygramDB	Speedup
Multi-word ("quantum physics")	104	2,566ms	0.09ms	27,600x
Medium-freq ("quantum")	1,961	1,874ms	0.28ms	6,700x
Low-freq ("algorithm")	2,498	507ms	0.42ms	1,200x
Rare ("fibonacci")	84	936ms	0.08ms	11,600x

CJK Search Latency (SORT id LIMIT 100)

CJK Search Latency (p50, log scale)

Query	Matches	MySQL	MygramDB	Speedup
日本	32,282	1,204ms	1.1ms	1,100x
東京	6,989	300ms	3.9ms	77x
科学	1,551	4.2ms	2.2ms	1.9x

COUNT Performance

COUNT Latency (p50, log scale)

Query Type	Count	MySQL	MygramDB	Speedup
Medium-freq ("quantum")	1,961	1,797ms	0.08ms	21,600x
Low-freq ("algorithm")	2,498	416ms	0.08ms	5,500x

Result Consistency (v1.5.0)

With verify_text: all, MygramDB produces exact match results with MySQL FULLTEXT:

Query	MySQL	MygramDB	Match
quantum	1,961	1,961	exact
algorithm	2,498	2,498	exact
日本	32,282	32,282	exact
科学	1,551	1,551	exact

Without verify_text, n-gram indexes may return false positives (e.g., "quantum" returns 58K instead of 1,961). This is inherent to n-gram tokenization and expected behavior.

Concurrent Throughput

Throughput — Queries per Second (higher is better)

Query: "algorithm", 10 seconds per level.

Connections	MySQL QPS	MygramDB QPS	MySQL p50	MygramDB p50
1	2	2,634	470ms	0.35ms
4	8	11,766	495ms	0.32ms

Memory Usage

Documents	Index	Documents + Text	Total RSS	Per 1M docs
1,100,000	813MB	1.54GB	2.53GB	~2.3GB

verify_text modes

• off (default): Lower memory, but may include n-gram false positives
• all: Stores document text for post-filter verification. Exact results at the cost of ~1.5GB additional memory for 1.1M docs.

Why These Numbers?

See Why MySQL FULLTEXT is Slow for architectural details, and Comparison for how MygramDB compares to Elasticsearch.