LiteSeed
Back

Technology

Performance at scale

A streaming architecture that generates millions of rows with \u2264 256 MB peak RAM.

Start Free

Performance benchmarks

Scale without memory constraints.

LiteSeed's chunk-based streaming architecture means generation speed and memory usage are independent of dataset size.

1M+

rows generated

≤ 256 MB

peak RAM at any scale

10K

rows per chunk

3

export formats

How it works

Chunk-based streaming

Generation, quality scoring and export all operate on chunks of 10,000 rows. No full dataset is ever held in memory.

  • \u2192CHUNK_SIZE = 10,000 rows constant
  • \u2192Each chunk generated, scored and written independently
  • \u2192Memory usage is O(chunk_size), not O(dataset_size)
  • \u2192Peak RAM ≤ 256 MB for 5M+ row runs

Streaming export writers

All three export formats use streaming writers that write rows as they are generated.

  • \u2192CSV: Node.js stream with csv-stringify
  • \u2192JSONL: line-by-line write, no buffering
  • \u2192Parquet: row-by-row via parquetjs-lite appendRow()
  • \u2192Export starts before generation completes

Mulberry32 PRNG performance

A high-performance, deterministic pseudo-random number generator optimised for JavaScript.

  • \u2192~500M samples/second in V8
  • \u219232-bit state — minimal memory footprint
  • \u2192Deterministic: same seed = same sequence
  • \u2192No external dependencies

Related