Why GPars runForkJoin is slow even when there are no forkOffChild()?

Question

This is the sequential version:

void f(long n) {
   for (int i=1; i<n-1; i++) {
      // do nothing     
   }
}

List result = []
(1..99999).each {
   f(it)
   result << it
}

It takes a few seconds to run the code.

void f(long n) {
   for (int i=1; i<n-1; i++) {
      // do nothing
   }
}

withPool {
   runForkJoin(1,99999) { a, b ->
      List result = []
      (a..b).each {
         f(it)
         result << it
      }
      return result
   }
}

The code above takes several minutes to finish. I haven't call any forkOffChild() or childrenResults() yet. I'm running this code in Windows and single core CPU with Intel Hyperthreading (2 logical CPU). Java Runtime.runtime.availableProcessors() returns 2.

I don't understand why the code that uses runForkJoin much slower than the sequential code (minutes versus seconds).

Answer 1

The runForJoin() method has no impact on performance in the code snippet. It is the withPool() method, that causes the slowdown. This is because the withPool() method adds several xxxParallel() methods to the Groovy Object's dynamic scope, which then slows down method resolution.

Annotating the f() method as @CompileStatic will give you the expected performance.