• Type: New Feature
    • Status: Closed
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 0.4.0
    • Component/s: None
    • Labels:


      This operator will allow sorting data that is larger than the size of memory by spilling to disk when necessary.

      Also as part of this jira, I will implement a new merge sort algorithm that will hopefully better utilize cluster resources than our current sort, which is based on Quicksort. The problem with quicksort is that we can't do any sorting until all of the batches have arrived. But this will often result in very low CPU utilization while the data is read from disk, followed by a period of very high CPU usage.

      The external sort will include sorting each batch individually when it arrives. In the case where no spills occur, it makes sense to take advantage of the fact that the batches are already sorted, but doing the an N-way merge, as is done when there are spills, is not the most effective way to do this, (according to some tests I have done). What will be done in this case, rather than an N-way merge using a heap, we will do a variation of natural merge sort, which amounts to essentially a staged, 2-way merge of the incoming (sorted) batches.


        1. DRILL-386.patch
          105 kB
          Steven Phillips
        2. DRILL-386.patch
          105 kB
          Steven Phillips
        3. DRILL-386.patch
          103 kB
          Steven Phillips
        4. DRILL-386.patch
          112 kB
          Steven Phillips



            • Assignee:
              sphillips Steven Phillips
              sphillips Steven Phillips
            • Votes:
              0 Vote for this issue
              2 Start watching this issue


              • Created: