fork join idle hands

build.sbt

scalaVersion := "2.10.1"

libraryDependencies += "jsr166" % "jsr166e" % "0.1" from "http://gee.cs.oswego.edu/dl/jsr166/dist/jsr166e.jar"

libraryDependencies += "jsr166" % "jsr166y" % "0.1" from "http://gee.cs.oswego.edu/dl/jsr166/dist/jsr166y.jar"

ConcurrentTest.scala

import annotation.tailrec
import java.util.concurrent.CountDownLatch

object ConcurrentTest {
  //
  // Uncomment on of the import lines below to choose a FJ implementation.
  //

  import scala.concurrent.forkjoin._ // The problem only shows up in the version of ForkJoin repackaged in Scala
  // import jsr166e._
  // import jsr166y._

  def banner(str: String) = {
    val line = "=" * 80
    println(s"\n$line\n$str\n$line")
  }

  def expensiveCalc() {
    @tailrec
    def loop(i: Int, accum: Double): Double =
      if (i == 0) accum else loop(i - 1, math.pow(math.sqrt(accum), 2.00001))
    loop(500000000, 42)
  }

  val N = Runtime.getRuntime.availableProcessors()
  val numTasks = N

  def nestedParallelismBench(pool: Pool, useFork: Boolean) {
    import pool.forkJoin
    val countDown = new CountDownLatch(numTasks)
    forkJoin.execute(new RecursiveAction {
      def compute() {
        (1 to numTasks) foreach { i =>
          val action = new RecursiveAction {
            def compute() {
              println(s"i = $i: ${Thread.currentThread.getName} ${forkJoin.toString.dropWhile(_ != '[')}")
              expensiveCalc()
              countDown.countDown()
            }
          }
          // See https://github.com/scala/scala/blob/2.10.1/src/library/scala/concurrent/impl/ExecutionContextImpl.scala#L118-#L120
          // The default ExecutionContext in Scala uses `fork` for nested parallelism.
          // This seems to leave a lot of cores idle. Why is this?
          if (useFork)
            action.fork()
          else
            forkJoin.execute(action)
        }
      }
    })
    countDown.await()
  }
  def nestedParallelismBenchRepeat(pool: Pool, useFork: Boolean) {
    (1 to 3).foreach { i =>
      banner(s"${pool.id}, useFork = $useFork, iteration $i")
      nestedParallelismBench(pool, useFork = useFork)
    }
  }

  case class Pool(forkJoin: ForkJoinPool, id: String)

  val pool = Pool(new ForkJoinPool(N), s"new ForkJoinPool($N)")

  def main(args: Array[String]) {
    nestedParallelismBenchRepeat(pool, useFork = false)
    nestedParallelismBenchRepeat(pool, useFork = true)
    nestedParallelismBenchRepeat(pool, useFork = false)
  }

}

s.c.forkjoin.log

// Using Scala 2.10.2.

================================================================================
new ForkJoinPool(8), useFork = false, iteration 1
================================================================================
i = 5: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 4, steals = 0, tasks = 0, submissions = 2]
i = 7: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 8, running = 3, steals = 0, tasks = 0, submissions = 1]
i = 6: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 0, tasks = 0, submissions = 2]
i = 4: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 7, active = 7, running = 6, steals = 0, tasks = 0, submissions = 2]
i = 2: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 4, active = 4, running = 4, steals = 0, tasks = 0, submissions = 2]
i = 3: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 5, active = 5, running = 5, steals = 0, tasks = 0, submissions = 2]
i = 1: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 4, active = 4, running = 3, steals = 0, tasks = 0, submissions = 2]
i = 8: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 2, steals = 0, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = false, iteration 2
================================================================================
i = 1: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 5, running = 2, steals = 9, tasks = 0, submissions = 5]
i = 2: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 3, steals = 9, tasks = 0, submissions = 5]
i = 3: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 4, steals = 9, tasks = 0, submissions = 5]
i = 4: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 8, running = 4, steals = 9, tasks = 0, submissions = 4]
i = 5: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 9, tasks = 0, submissions = 3]
i = 6: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 9, tasks = 0, submissions = 2]
i = 7: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 9, tasks = 0, submissions = 1]
i = 8: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 8, running = 8, steals = 9, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = false, iteration 3
================================================================================
i = 1: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 2, steals = 18, tasks = 0, submissions = 7]
i = 4: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 18, tasks = 0, submissions = 3]
i = 5: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 18, tasks = 0, submissions = 3]
i = 3: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 18, tasks = 0, submissions = 3]
i = 2: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 18, tasks = 0, submissions = 3]
i = 6: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 18, tasks = 0, submissions = 2]
i = 7: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 18, tasks = 0, submissions = 0]
i = 8: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 18, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = true, iteration 1
================================================================================
i = 8: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 2, steals = 27, tasks = 7, submissions = 0]
i = 1: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 5, submissions = 0]
i = 2: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 5, submissions = 0]
i = 3: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 4, submissions = 0]
i = 7: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 3, submissions = 0]
i = 4: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 2, submissions = 0]
i = 5: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 1, submissions = 0]
i = 6: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 27, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = true, iteration 2
================================================================================
i = 8: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 1, steals = 33, tasks = 7, submissions = 0]
i = 1: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 5, submissions = 0]
i = 2: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 5, submissions = 0]
i = 7: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 4, submissions = 0]
i = 3: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 3, submissions = 0]
i = 4: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 2, submissions = 0]
i = 6: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 1, submissions = 0]
i = 5: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 33, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = true, iteration 3
================================================================================
i = 8: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 1, steals = 39, tasks = 7, submissions = 0]
i = 1: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 5, submissions = 0]
i = 2: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 5, submissions = 0]
i = 3: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 4, submissions = 0]
i = 7: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 3, submissions = 0]
i = 4: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 2, submissions = 0]
i = 5: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 1, submissions = 0]
i = 6: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 3, running = 3, steals = 39, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = false, iteration 1
================================================================================
i = 2: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 46, tasks = 0, submissions = 6]
i = 1: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 6, running = 2, steals = 46, tasks = 0, submissions = 4]
i = 4: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 46, tasks = 0, submissions = 2]
i = 3: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 46, tasks = 0, submissions = 2]
i = 5: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 46, tasks = 0, submissions = 2]
i = 7: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 46, tasks = 0, submissions = 1]
i = 6: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 46, tasks = 0, submissions = 2]
i = 8: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 46, tasks = 0, submissions = 0]

================================================================================
new ForkJoinPool(8), useFork = false, iteration 2
================================================================================
i = 6: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 8, running = 4, steals = 55, tasks = 0, submissions = 1]
i = 3: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 8, active = 8, running = 6, steals = 55, tasks = 0, submissions = 0]
i = 4: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 55, tasks = 0, submissions = 0]
i = 2: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 55, tasks = 0, submissions = 0]
i = 7: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 8, active = 8, running = 3, steals = 55, tasks = 0, submissions = 0]
i = 1: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 55, tasks = 0, submissions = 0]
i = 8: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 5, steals = 55, tasks = 0, submissions = 0]
i = 5: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 55, tasks = 0, submissions = 1]

================================================================================
new ForkJoinPool(8), useFork = false, iteration 3
================================================================================
i = 2: ForkJoinPool-1-worker-7 [Running, parallelism = 8, size = 8, active = 7, running = 4, steals = 64, tasks = 0, submissions = 3]
i = 7: ForkJoinPool-1-worker-5 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 64, tasks = 0, submissions = 1]
i = 4: ForkJoinPool-1-worker-2 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 64, tasks = 0, submissions = 1]
i = 8: ForkJoinPool-1-worker-1 [Running, parallelism = 8, size = 8, active = 8, running = 3, steals = 64, tasks = 0, submissions = 0]
i = 3: ForkJoinPool-1-worker-8 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 64, tasks = 0, submissions = 1]
i = 6: ForkJoinPool-1-worker-6 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 64, tasks = 0, submissions = 1]
i = 5: ForkJoinPool-1-worker-3 [Running, parallelism = 8, size = 8, active = 8, running = 7, steals = 64, tasks = 0, submissions = 1]
i = 1: ForkJoinPool-1-worker-4 [Running, parallelism = 8, size = 8, active = 6, running = 4, steals = 64, tasks = 0, submissions = 3]

Process finished with exit code 0