Nio vs Io File Absence performance

comments.txt

Apparently the "old" File API is about 3 times as fast here on my mac. 
I have not tested it on other OS. Note that the GC stats are misleading: 
I got confused why the old File API creates more garbage. It does not.
Check the "_norm" values. Because it has about 3x the iterations,
its total gc time and count is of course higher.

FileAccess.java

import java.io.File;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Scope;
import org.openjdk.jmh.annotations.State;
// uses counter to ensure we get a cache miss from whatever might be caching
@State(Scope.Thread)
public class FileAccess {

  int i = 0;

  @Benchmark
  public boolean nio() {
    return Files.exists(Paths.get("/Users/fabian/work/temp/", "nio" + i++));
  }

  @Benchmark
  public boolean io() {
    return new File("/Users/fabian/work/temp/", "io" + i++).exists();
  }
}

gc.txt

Benchmark                                         Mode  Cnt       Score       Error   Units
FileAccess.io                                    thrpt   20  243762.181 ± 11680.500   ops/s
FileAccess.io:·gc.alloc.rate                     thrpt   20     182.197 ±     8.744  MB/sec
FileAccess.io:·gc.alloc.rate.norm                thrpt   20     784.005 ±     0.010    B/op
FileAccess.io:·gc.churn.PS_Eden_Space            thrpt   20     180.112 ±    16.788  MB/sec
FileAccess.io:·gc.churn.PS_Eden_Space.norm       thrpt   20     774.989 ±    62.758    B/op
FileAccess.io:·gc.churn.PS_Survivor_Space        thrpt   20       0.109 ±     0.031  MB/sec
FileAccess.io:·gc.churn.PS_Survivor_Space.norm   thrpt   20       0.472 ±     0.141    B/op
FileAccess.io:·gc.count                          thrpt   20     108.000              counts
FileAccess.io:·gc.time                           thrpt   20      48.000                  ms
FileAccess.nio                                   thrpt   20   63042.097 ±  4388.955   ops/s
FileAccess.nio:·gc.alloc.rate                    thrpt   20     132.819 ±    10.539  MB/sec
FileAccess.nio:·gc.alloc.rate.norm               thrpt   20    2208.994 ±    70.221    B/op
FileAccess.nio:·gc.churn.PS_Eden_Space           thrpt   20     134.349 ±    19.286  MB/sec
FileAccess.nio:·gc.churn.PS_Eden_Space.norm      thrpt   20    2231.454 ±   262.013    B/op
FileAccess.nio:·gc.churn.PS_Survivor_Space       thrpt   20       0.086 ±     0.023  MB/sec
FileAccess.nio:·gc.churn.PS_Survivor_Space.norm  thrpt   20       1.429 ±     0.395    B/op
FileAccess.nio:·gc.count                         thrpt   20      72.000              counts
FileAccess.nio:·gc.time                          thrpt   20      32.000                  ms

perf.txt

# JMH 1.10.1 (released 11 days ago)
# VM invoker: /Library/Java/JavaVirtualMachines/jdk1.8.0_40.jdk/Contents/Home/jre/bin/java
# VM options: <none>
# Warmup: 5 iterations, 1 s each
# Measurement: 20 iterations, 1 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark: FileAccess.io

Result "io":
  267340.600 ±(99.9%) 12996.163 ops/s [Average]
  (min, avg, max) = (227755.704, 267340.600, 282358.171), stdev = 14966.401
  CI (99.9%): [254344.436, 280336.763] (assumes normal distribution)


# JMH 1.10.1 (released 11 days ago)
# VM invoker: /Library/Java/JavaVirtualMachines/jdk1.8.0_40.jdk/Contents/Home/jre/bin/java
# VM options: <none>
# Warmup: 5 iterations, 1 s each
# Measurement: 20 iterations, 1 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark: FileAccess.nio

Result "nio":
  77926.363 ±(99.9%) 5559.603 ops/s [Average]
  (min, avg, max) = (66773.390, 77926.363, 83652.838), stdev = 6402.447
  CI (99.9%): [72366.760, 83485.966] (assumes normal distribution)


# Run complete. Total time: 00:00:50

Benchmark        Mode  Cnt       Score       Error  Units
FileAccess.io   thrpt   20  267340.600 ± 12996.163  ops/s
FileAccess.nio  thrpt   20   77926.363 ±  5559.603  ops/s