JSON Performance

borer comes with a fast JSON parser optimized for modern superscalar 64-bit CPUs, which performs quite well when compared to other popular JSON libraries in the scala eco-system when it comes to transforming raw JSON-bytes into a case-class based data model.

A benchmark against 18 real-world JSON data examples from a diverse set of sources shows that borer’s JSON parsing infrastructure outperforms Circe (for example) by more than a factor of two, on average:

This is the benchmark setup:

Task

Decode 18 files from raw JSON bytes into a respective case-class based data model generated with https://transform.now.sh/json-to-scala-case-class/

JMH Parameters

-wi 10 -i 10 -t 4 -f 4

JVM

Java HotSpot(TM) 64-Bit GraalVM EE 19.2.1

Memory Settings

-Xms4096m -Xmx4096m

Scala Version

2.12.10

Operating System

Ubuntu 18.04.2 LTS

CPU

2 x Intel Xeon E5-2665 2.4 - 3.1 GHz (Octa-core)

The benchmarks SBT sub project contains everything you need to run the benchmarks yourself. This SBT command will produce a JSON file holding all results:

jmh:run io.bullet.borer.benchmarks.* -wi 10 -i 10 -t 4 -f 4 -foe true -rf json -rff results/my_results.json  

Since interpreting the results can be a bit tedious borer comes with some simple analysis logic that can be invoked with the benchmarkResults SBT command. It will show you a synthesis of all benchmark results currently living in the benchmarks/results folder.