A micro-benchmark to compare the throughput of two sorting methods provided by the java.util.Arrays class,
parallelSort and sort. The benchmark generates an array of n random numbers. The
resulting unsorted array is then sorted by each method.
Tested with: Maven 3.8.4 and Amazon Corretto JDK 17
- Step 1:
$ git clone git://github.com/rharri/streams-benchmark.git - Step 2:
$ cd array-sort-bench - Step 3:
$ mvn clean verify - Step 4:
$ java -jar target/benchmarks.jar
| Machine | |
|---|---|
| Operating System | Ubuntu 20.04.3 LTS |
| CPU | 3.7GHz 6-Core AMD Ryzen 5 5600X |
| RAM | 32 GB 3200 MHz DDR4 |
| JVM | OpenJDK 64-Bit Server VM Corretto-17.0.2.8.1 (build 17.0.2+8-LTS, mixed mode, sharing) |
The size represents the number of random numbers generated, and thus number of elements to be sorted by each
method. In this case: 1 Hundred, 1 Thousand, 4 Thousand, 10 Thousand, 100 Thousand, 1 Million, and 10 Million.
| Benchmark | (size) | Mode | Cnt | Score | Error | Units |
|---|---|---|---|---|---|---|
| testParallelSort | 100 | thrpt | 8 | 2065213.082 | ± 597319.145 | ops/s |
| testParallelSort | 1000 | thrpt | 8 | 135466.730 | ± 6844.356 | ops/s |
| testParallelSort | 4000 | thrpt | 8 | 17151.777 | ± 1572.506 | ops/s |
| testParallelSort | 10000 | thrpt | 8 | 7061.721 | ± 2213.384 | ops/s |
| testParallelSort | 100000 | thrpt | 8 | 1371.719 | ± 18.590 | ops/s |
| testParallelSort | 1000000 | thrpt | 8 | 159.374 | ± 15.976 | ops/s |
| testParallelSort | 10000000 | thrpt | 8 | 14.856 | ± 0.066 | ops/s |
| testSort | 100 | thrpt | 8 | 1941952.830 | ± 392573.519 | ops/s |
| testSort | 1000 | thrpt | 8 | 128412.026 | ± 4753.738 | ops/s |
| testSort | 4000 | thrpt | 8 | 17166.564 | ± 2639.573 | ops/s |
| testSort | 10000 | thrpt | 8 | 3936.256 | ± 7.219 | ops/s |
| testSort | 100000 | thrpt | 8 | 288.131 | ± 3.097 | ops/s |
| testSort | 1000000 | thrpt | 8 | 24.575 | ± 0.336 | ops/s |
| testSort | 10000000 | thrpt | 8 | 2.036 | ± 0.166 | ops/s |
thrpt = Throughput, ops/s = Operations per second
The numbers above are just data. To gain reusable insights, you need to follow up on why the numbers are the way they are. Use profilers (see -prof, -lprof), design factorial experiments, perform baseline and negative tests that provide experimental control, make sure the benchmarking environment is safe on JVM/OS/HW level, ask for reviews from the domain experts. Do not assume the numbers tell you what you want them to tell.