rubensworks / test-performance-json-parse-stream

basic measurements that measure the performance and continuous memory usage of several streaming JSON parsers for JavaScript

Performance measurements of streaming JSON parsers

This repo contains some basic measurements that measure the performance and continuous memory usage of several streaming JSON parsers for JavaScript.

Note that this repo does not aim to be a fully-fledged benchmark. It's just a small test for myself.

Considered metrics:

• Execution time
• Memory usage
• Robustness (unit tests)
• Usage (number of dependents)
• Maintained (last release)

Tested libraries

• stream-json
• jsonparse
• SAX variant of jsonparse
• clarinet

Executing these measurements

First, clone this repo.

Then, install and execute it with some JSON file:

$ npm install
$ node --expose-gc index.js some-file.json

In my experiments, the input JSON file (144,2MB) was created as follows:

$ npm install -g @comunica/actor-init-sparql
$ comunica-sparql https://fragments.dbpedia.org/2016-04/en 'construct where { ?s ?p ?o } limit 1000000' -t application/ld+json > dbpedia-1m.jsonld

Results

Results obtained by executing on MacBook Pro (13-inch, 2020, Four Thunderbolt 3 ports); 2,3 GHz Quad-Core Intel Core i7; 16 GB 3733 MHz LPDDR4

Library	Execution time	Memory	Robustness	Usage	Maintained	Notes
stream-json	3.626s	65MB	High	139	June 2021	Extensive documentation
jsonparse	2.791s	384MB	High	175	2017	Accumulates memory!
SAX variant of jsonparse	1.431s	72MB	Low	0	/	Not a proper npm package
clarinet	2.213s	67MB	High	33	2019

Conclusion

Performance-wise, the SAX variant of jsonparse is the clear winner. However, since it is unmaintained, and no tests are available, it is not recommended.

While jsonparse itself is also pretty fast, it suffers from memory issues for large streams (as it still accumulates JSON in memory). So this is also a no-go.

stream-json and clarinet both have acceptable performance and memory usage. While stream-json is slower than clarinet, it seems to be much better maintained, and has extensive documentation and a broader usage.

Therefore, even though stream-json is slower, it may be considered the best library available at the moment for parsing JSON in a streaming manner.

License

This software is written by Ruben Taelman.

This code is released under the MIT license.