// Performance

Benchmarks

Real performance numbers from DRAVN's benchmark suite. Tested with 1M rows of synthetic USA weather data across PostgreSQL, SQLite, and Kafka on a single machine.

Node v24.12PostgreSQL 16SQLite (better-sqlite3)Kafka 7.6 (KRaft)Windows 11

311x

ELT vs In-Process

1,332x

Batch vs Row-by-Row

300K

Rows/sec Streaming

280K

SQLite Read rows/s

20K

Kafka msg/sec

Execution Mode Comparison

100,000 rows — aggregate by state and year. Same source, same target, same transform. The only difference is where the work happens.

// BASELINE

600

rows / second

166.72s for 100K rows

// RECOMMENDED

186,342

rows / second

0.54s for 100K rows

ELT is 311x faster — data never leaves the database.

PostgreSQL Batch Write Performance

100,000 rows written to PostgreSQL with different batch sizes. Multi-row INSERT with transactions.

Batch Size	Duration	Rows/sec	Speedup
1 (row-by-row)	47.38s	21	1x
50	89.22s	1,121	53x
100	45.31s	2,207	105x
500	8.03s	12,446	593x
1,000	5.33s	18,747	893x
5,000	3.57s	27,977	1,332x

Source Read Speed

100,000 rows read from different source types into memory.

SQLite280,218 rows/s

JSON file147,111 rows/s

PostgreSQL103,551 rows/s

CSV file91,181 rows/s

Streaming 1M Rows

1,000,000 rows streamed with backpressure at different chunk sizes. Memory stays flat — only one chunk in memory at a time.

Chunk Size	Duration	Rows/sec	Peak Memory
100	109.63s	9,121	580 MB
500	34.43s	29,047	579 MB
1,000	18.20s	54,942	586 MB
5,000	3.33s	300,341	586 MB

Memory stays at ~580 MB regardless of chunk size — backpressure works.

SQLite Performance (Lite Mode)

SQLite with WAL mode — the foundation of DRAVN Lite. No external database needed.

289,233

READ rows/sec

168,947

WRITE rows/sec

244,173

CONCURRENT R+W

Write Batch Size Comparison

Batch 112,894 rows/s

Batch 50131,852 rows/s

Batch 100144,934 rows/s

Batch 500161,718 rows/s

Batch 1,000168,947 rows/s

Batch 5,000133,138 rows/s

Kafka Streaming

Apache Kafka 7.6 in KRaft mode. Produce, consume, and end-to-end pipeline throughput.

Test	Messages	msg/sec
Produce batch 50	10,000	19,996
Produce batch 1000	10,000	16,919
Consume	10,000	2,192
E2E produce+consume	10,000	2,642
Kafka → PostgreSQL	10,000	2,414

Concurrent Pipeline Stress

Sequential (3x 10K)

10,658

rows / second

Parallel (3x 10K)

23,349

rows / second

Parallel pipelines: 2.2x faster than sequential.

NOTE

Methodology

All benchmarks run on a single machine using DRAVN's built-in benchmark suite (npm run benchmark:full). Data is 1M rows of synthetic USA weather station readings (14 columns, ~258 MB JSON).

PostgreSQL 16 and Redis 7 run in Podman containers. SQLite uses better-sqlite3 with WAL mode. Kafka 7.6 runs in KRaft mode (no ZooKeeper). Timing uses performance.now(); memory via process.memoryUsage().heapUsed.

Numbers will vary by hardware, network, and data shape. Run the benchmark suite on your own infrastructure for accurate results.