Back

Explore Courses Blog Tutorials Interview Questions
0 votes
1 view
in Big Data Hadoop & Spark by (11.4k points)

In below screenshot of Spark admin running on port 8080 :

enter image description here

The "Shuffle Read" & "Shuffle Write" parameters are always empty for this code :

import org.apache.spark.SparkContext;

object first {
  println("Welcome to the Scala worksheet")

  val conf = new org.apache.spark.SparkConf()
    .setMaster("local")
    .setAppName("distances")
    .setSparkHome("C:\\spark-1.1.0-bin-hadoop2.4\\spark-1.1.0-bin-hadoop2.4")
    .set("spark.executor.memory", "2g")
  val sc = new SparkContext(conf)

  def euclDistance(userA: User, userB: User) = {

    val subElements = (userA.features zip userB.features) map {
      m => (m._1 - m._2) * (m._1 - m._2)
    }
    val summed = subElements.sum
    val sqRoot = Math.sqrt(summed)

    println("value is" + sqRoot)
    ((userA.name, userB.name), sqRoot)
  }

  case class User(name: String, features: Vector[Double])

  def createUser(data: String) = {

    val id = data.split(",")(0)
    val splitLine = data.split(",")

    val distanceVector = (splitLine.toList match {
      case h :: t => t
    }).map(m => m.toDouble).toVector

    User(id, distanceVector)

  }

  val dataFile = sc.textFile("c:\\data\\example.txt")
  val users = dataFile.map(m => createUser(m))
  val cart = users.cartesian(users) //
  val distances = cart.map(m => euclDistance(m._1, m._2))
  //> distances  : org.apache.spark.rdd.RDD[((String, String), Double)] = MappedR
  //| DD[4] at map at first.scala:46
  val d = distances.collect //

  d.foreach(println) //> ((a,a),0.0)
  //| ((a,b),0.0)
  //| ((a,c),1.0)
  //| ((a,),0.0)
  //| ((b,a),0.0)
  //| ((b,b),0.0)
  //| ((b,c),1.0)
  //| ((b,),0.0)
  //| ((c,a),1.0)
  //| ((c,b),1.0)
  //| ((c,c),0.0)
  //| ((c,),0.0)
  //| ((,a),0.0)
  //| ((,b),0.0)
  //| ((,c),0.0)
  //| ((,),0.0)

}


Why are "Shuffle Read" & "Shuffle Write" fields empty ? 

1 Answer

0 votes
by (32.3k points)

Shuffling is the process of data transfer between stages or can be determined as a process where the reallocation of data between multiple Spark stages. "Shuffle Write" is actually meant as the sum of all written serialized data on all executors before transmitting (normally at the end of a stage) and "Shuffle Read" means the sum of read serialized data on all executors at the beginning of a stage.

Your program has only one stage, triggered by the "collect" operation. No shuffling is required, because you have only a bunch of consecutive map operations which are pipelined in one Stage.

Try to take a look at these slides: http://de.slideshare.net/colorant/spark-shuffle-introduction

Welcome to Intellipaat Community. Get your technical queries answered by top developers!

28.4k questions

29.7k answers

500 comments

94k users

Browse Categories

...