Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-36420

Use `isEmpty` to improve performance in Pregel's superstep

    XMLWordPrintableJSON

Details

    • Improvement
    • Status: Resolved
    • Minor
    • Resolution: Fixed
    • 3.3.0
    • 3.3.0
    • GraphX
    • None

    Description

      When I was running `Graphx.connectedComponents` with 20+ billion vertices and edges, I found that count is very slow.

      object Pregel extends Logging {
        ...
        def apply[VD: ClassTag, ED: ClassTag, A: ClassTag] (...): Graph[VD, ED] = {
          ...
          // Maybe messages.isEmpty() is better than messages.count()
          var activeMessages = messages.count()
      
          // Loop
          var prevG: Graph[VD, ED] = null
          var i = 0
          while (activeMessages > 0 && i < maxIterations) {
            ...
            activeMessages = messages.count()
            ...
          }
          ...
          g
        } // end of apply
      
      } // end of class Pregel
      

      Maybe we only need an action operator here and active-messages are not empty, so we don’t need to use count, it’s better to use isEmpty. I verified it and it worked very well.

       

      Attachments

        Activity

          People

            xiepengjie xiepengjie
            xiepengjie xiepengjie
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: