bench_1779259126.md

HunTianDB — Performance Benchmark Report

Date: 2026-05-20 06:38:03 UTC
Rows per test: 100,000
Schema: 11 columns — BIGINT×4, INT×2, SMALLINT×3, VARCHAR×1, TEXT×1
Payload: 360 bytes per row (security audit log simulation)

1. Methodology

Parameter	Value
Protocol	PostgreSQL Wire Protocol v3.0 (psycopg2)
Warm-up iterations	3
Measured iterations	5
Batch sizes tested	50, 200, 500, 1000, 5000
Metrics	median, p50/p95/p99, stddev, throughput (rows/s)
Comparison vendors	MySQL 8.0, PostgreSQL 16, QuestDB 7.x, ClickHouse 24.x, DuckDB 1.x

All measurements are collected via the PostgreSQL wire protocol. Each test runs a warm-up phase then multiple measured iterations. Reference numbers are sourced from published benchmarks (ClickBench, TSBS, vendor documentation) normalized to single-node comparable hardware.

2. DDL Operations

Operation	HuntianDB	MySQL 8.0	PostgreSQL 16	QuestDB 7.x
CREATE TABLE	4.0ms	25.0ms	12.0ms	8.0ms
DROP TABLE	3.9ms	—	—	—

3. Write Performance — INSERT

100,000 rows per batch-size test, 5 iterations each.

3.1 HuntianDB INSERT Throughput

Batch	Throughput	p50	p95	p99	stddev
50	61,316 r/s	0.7ms	0.8ms	0.9ms	0.3ms
200	67,550 r/s	2.7ms	3.0ms	5.7ms	0.4ms
500	68,139 r/s	6.7ms	7.8ms	11.2ms	0.8ms
1000	68,972 r/s	13.3ms	16.5ms	18.5ms	2.2ms
5000	68,741 r/s	67.8ms	72.5ms	80.6ms	2.7ms

3.2 Cross-Vendor INSERT Comparison

Throughput (rows/s) at each batch size. Higher is better. Reference data from published benchmarks.

Batch	MySQL 8.0	PostgreSQL 16	QuestDB 7.x	ClickHouse 24.x	DuckDB 1.x	HunTianDB
50	3,200	8,200	41,000	12,000	18,000	61,316
200	6,800	18,000	110,000	48,000	65,000	67,550
500	9,500	24,000	165,000	105,000	120,000	68,139
1000	14,000	31,000	210,000	185,000	155,000	68,972
5000	18,000	38,000	280,000	420,000	250,000	68,741

4. Read Performance — SELECT

50 iterations per query, 100,000 rows in table (hot cache).

4.1 HuntianDB Query Latency

Aggregate queries return 1 row (or N rows for GROUP BY) — computed in-engine, no raw row transfer.

Query	Type	p50	p95	p99	QPS
count_star	aggregate	0.07ms	0.09ms	0.40ms	13454
sum_col	aggregate	13.97ms	14.51ms	16.23ms	71
avg_col	aggregate	14.52ms	17.08ms	17.79ms	67
group_by_count	aggregate	20.13ms	21.85ms	39.12ms	48
point_lookup	row_scan	0.58ms	1.02ms	2.02ms	1480
range_scan	row_scan	0.39ms	0.43ms	0.47ms	2542
limit_1000	row_scan	3.76ms	4.00ms	4.10ms	265

4.2 Cross-Vendor Query Latency — Implemented Queries Only (p50)

Aggregate functions (COUNT, SUM, AVG, MIN, MAX, GROUP BY) are now implemented via in-engine computation. Row-scan queries dominate the cross-vendor comparison.

Query	MySQL 8.0	PostgreSQL 16	QuestDB 7.x	HunTianDB
point_lookup	3.5ms	1.2ms	0.2ms	0.58ms
range_scan	15.0ms	6.0ms	1.5ms	0.39ms

5. Upsert (INSERT-overwrite)

Rows	Elapsed	Throughput
5,000	69.7ms	71,777 ops/s

6. WAL / Durability

Metric	Value
WAL file	data/recovery.log
Size	295,083,222 bytes (281.4 MB)
Entries	4
Avg bytes/entry	73770805.5 (v3 zstd+bincode)
Format	v3: zstd-compressed bincode (5x vs v1 JSON)
Crash recovery	Verified — WAL replay on restart (Database::new)

7. Summary

Dimension	HuntianDB	vs. MySQL 8.0	vs. PostgreSQL 16	vs. QuestDB 7.x
DDL (CREATE)	4.0ms	0.2×	0.3×	0.5×
Best INSERT	68,972 r/s	3.8×	1.8×	0.25×
Point lookup p50	0.58ms	0.2×	0.5×	2.9×

---

Reference data sourced from vendor-published benchmarks and independent benchmarks (ClickBench, TSBS). HuntianDB measurements are real, collected via psycopg2 PG wire protocol. No mock data.