Trying to benchmark Java 8 parallelSetAll() vs setAll()
This question already has an answer here:
Using jmh, I get the following results (size is the length of the array). The score is the runtime per call, in micro seconds (smaller = better).
Benchmark (size) Mode Cnt Score Error Units
SO34929316.parallelSetAll 1 avgt 20 0.077 ± 0.003 us/op
SO34929316.parallelSetAll 1000 avgt 20 9.935 ± 0.478 us/op
SO34929316.parallelSetAll 100000 avgt 20 56.024 ± 7.008 us/op
SO34929316.parallelSetAll 1000000 avgt 20 518.495 ± 75.331 us/op
SO34929316.parallelSetAll 10000000 avgt 20 5640.574 ± 139.324 us/op
SO34929316.setAll 1 avgt 20 0.016 ± 0.001 us/op
SO34929316.setAll 1000 avgt 20 1.257 ± 0.023 us/op
SO34929316.setAll 100000 avgt 20 116.760 ± 3.116 us/op
SO34929316.setAll 1000000 avgt 20 1168.868 ± 42.153 us/op
SO34929316.setAll 10000000 avgt 20 12347.399 ± 766.931 us/op
As you would expect, parallelSetAll is faster for larger arrays and slower for smaller arrays. However the speedup factor is only ~ x2 according to the test (ran on Win 8.1 / Xeon 2630 v2 / 6 cores (x2)).
For reference, the code:
@State(Scope.Thread)
@BenchmarkMode(Mode.AverageTime)
public class SO34929316 {
@Param({"1", "1000", "100000", "1000000", "10000000"}) int size;
long[] array;
@Setup(Level.Iteration)
public void setup(){
array = new long[size];
}
@Benchmark
public void setAll(Blackhole bh) {
Arrays.setAll(array, n -> n * n * 11 + n * 7);
bh.consume(array);
}
@Benchmark
public void parallelSetAll(Blackhole bh) {
Arrays.parallelSetAll(array, n -> n * n * 11 + n * 7);
bh.consume(array);
}
}
上一篇: 为什么循环比while循环更快