Java & Kotlin Micro Benchmarks

Collection of small micro benchmarks of Java and Kotlin snippets.

Also check out my Kotlin Multiplatform Benchmark project on GitHub:
👉 kotlin-multiplatform-benchmark.

Benchmark tool suite for performance tests based on 💿 JMH Java Microbenchmark Harness.
To create your own micro-benchmark, you can check out the tutorial on Baeldung:
📖 Microbenchmarking with Java

🚀 Start

Start with Maven

Build the project:

mvn clean verify

Then run the generated jar:

java -jar target/benchmarks.jar

Start with IDEA

To run the benchmark within your IDE, there is a Runner for each benchmark:
Class: de.tfr.benchmark.numbersuffix.NumberSuffixBenchmarkRunner
Run Config: ▶ PidMappingsBenchmarkRunner

Start with JMH Plugin

Alternatively you can use the IntelliJ Plugin:

💿 JMH Java Microbenchmark Harness

This allows starting tests like other JUnit tests:

📊 Benchmark Results

Puid mappings

NumberSuffixBenchmark
Benchmark for getting the 3 digit suffix of a number. So simply cut of the last 3 digits of a number. Sample input data: 1000999
Sample output data: 1000 Sample output type: int The benchmark compares the cutting number method with String.substring() vs` a simple division.

Benchmark                      (iterations)  Mode  Cnt   Score    Error  Units
NumberSuffixBenchmark.getPrefixByString            10  avgt    3   0,001 ±  0,001  ms/op
NumberSuffixBenchmark.getPrefixByString           100  avgt    3   0,010 ±  0,001  ms/op
NumberSuffixBenchmark.getPrefixByString          1000  avgt    3   0,104 ±  0,014  ms/op
NumberSuffixBenchmark.getPrefixByString         10000  avgt    3   0,937 ±  0,368  ms/op
NumberSuffixBenchmark.getPrefixByString            10  avgt    3  ≈ 10⁻⁵           ms/op
NumberSuffixBenchmark.getPrefixByMath             100  avgt    3  ≈ 10⁻⁴           ms/op
NumberSuffixBenchmark.getPrefixByMath            1000  avgt    3   0,001 ±  0,001  ms/op
NumberSuffixBenchmark.getPrefixByMath           10000  avgt    3   0,010 ±  0,006  ms/op

The getSuffixByMath is more than 93 times faster, than the String operations!
In a project, parsing of 5181276 ids for one DB entity, decreased from 480ms to just 5ms.
It was possible to save 213 Seconds for all entity (2304183145 ids)!

Custom Json Parser

CustomJsonParserBenchmark
A custom parser for JSON int arrays vs kotlinSerialization. The KotlinSerializationJsonDecoder is used internally by the Spring WebClient during the execution of an HTTP request.
Sample input data: [100,101]
Sample output data: primitive list of integer.

Benchmark                                (iterations)  Mode  Cnt     Score      Error  Units
NumberSuffixBenchmark.customParseJson               10  avgt    3     0,007 ±    0,001  ms/op
NumberSuffixBenchmark.customParseJson              100  avgt    3     0,461 ±    0,120  ms/op
NumberSuffixBenchmark.customParseJson             1000  avgt    3    50,073 ±    6,391  ms/op
NumberSuffixBenchmark.customParseJson            10000  avgt    3  5065,790 ± 1732,779  ms/op
NumberSuffixBenchmark.kotlinSerialization           10  avgt    3     0,019 ±    0,002  ms/op
NumberSuffixBenchmark.kotlinSerialization          100  avgt    3     0,961 ±    0,283  ms/op
NumberSuffixBenchmark.kotlinSerialization         1000  avgt    3    82,318 ±   44,921  ms/op
NumberSuffixBenchmark.kotlinSerialization        10000  avgt    3  7905,643 ± 3148,923  ms/op

The custom Integer parser is 35,92% faster.

Custom Integer Parser

IntegerParserBenchmark
Custom integer primitive number parser vs Integer.parseInt(..).

Benchmark                              (iterations)  Mode  Cnt   Score    Error  Units
IntegerParserBenchmark.customParseInt            10  avgt    3  ≈ 10⁻⁴           ms/op
IntegerParserBenchmark.customParseInt           100  avgt    3   0,001 ±  0,001  ms/op
IntegerParserBenchmark.customParseInt          1000  avgt    3   0,015 ±  0,003  ms/op
IntegerParserBenchmark.customParseInt         10000  avgt    3   0,162 ±  0,065  ms/op
IntegerParserBenchmark.parseInt                  10  avgt    3  ≈ 10⁻³           ms/op
IntegerParserBenchmark.parseInt                 100  avgt    3   0,004 ±  0,001  ms/op
IntegerParserBenchmark.parseInt                1000  avgt    3   0,042 ±  0,050  ms/op
IntegerParserBenchmark.parseInt               10000  avgt    3   0,280 ±  1,968  ms/op

The custom customParseInt ist 72,82 % faster than Integer.parseInt().
But the absolut benefit might be small, because even 10000 Iteration tage only 0,132 ms.

🍴 Fork

This project is based on the jmh Maven template:

 mvn archetype:generate \
  -DinteractiveMode=false \
  -DarchetypeGroupId=org.openjdk.jmh \
  -DarchetypeArtifactId=jmh-java-benchmark-archetype \
  -DgroupId=de.tfr.benchmark \
  -DartifactId=redgiant-bench \
  -Dversion=1.0

TobseF/JaKoMicrobenchmarks