Skip to content

[C++] Large regression in single-key table sort performance #36176

Closed

Description

@pitrou

Describe the bug, including details regarding any error messages, version, and platform.

Following GH-33206, a large performance regression appeared in single-key table sorting:

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Regressions: (11)
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
                                   benchmark           baseline         contender  change %                                                                                                                                                                                                                                       counters
     ArraySortIndicesInt64Narrow/32768/10000      2.683 GiB/sec     2.165 GiB/sec   -19.296                                   {'family_index': 0, 'per_family_instance_index': 0, 'run_name': 'ArraySortIndicesInt64Narrow/32768/10000', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 59563, 'null_percent': 0.01}
TableSortIndicesInt64Narrow/1048576/100/1/32  16.175M items/sec 10.063M items/sec   -37.786 {'family_index': 12, 'per_family_instance_index': 9, 'run_name': 'TableSortIndicesInt64Narrow/1048576/100/1/32', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 11, 'chunks': 32.0, 'columns': 1.0, 'null_percent': 1.0}
  TableSortIndicesInt64Narrow/1048576/4/1/32  19.455M items/sec 11.938M items/sec   -38.639 {'family_index': 12, 'per_family_instance_index': 10, 'run_name': 'TableSortIndicesInt64Narrow/1048576/4/1/32', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 13, 'chunks': 32.0, 'columns': 1.0, 'null_percent': 25.0}
  TableSortIndicesInt64Narrow/1048576/0/1/32  18.017M items/sec 10.212M items/sec   -43.320  {'family_index': 12, 'per_family_instance_index': 11, 'run_name': 'TableSortIndicesInt64Narrow/1048576/0/1/32', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 12, 'chunks': 32.0, 'columns': 1.0, 'null_percent': 0.0}
   TableSortIndicesInt64Narrow/1048576/4/1/4  35.396M items/sec 15.609M items/sec   -55.901   {'family_index': 12, 'per_family_instance_index': 22, 'run_name': 'TableSortIndicesInt64Narrow/1048576/4/1/4', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 25, 'chunks': 4.0, 'columns': 1.0, 'null_percent': 25.0}
 TableSortIndicesInt64Narrow/1048576/100/1/4  34.966M items/sec 12.808M items/sec   -63.370  {'family_index': 12, 'per_family_instance_index': 21, 'run_name': 'TableSortIndicesInt64Narrow/1048576/100/1/4', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 23, 'chunks': 4.0, 'columns': 1.0, 'null_percent': 1.0}
   TableSortIndicesInt64Narrow/1048576/0/1/4  36.230M items/sec 13.005M items/sec   -64.103    {'family_index': 12, 'per_family_instance_index': 23, 'run_name': 'TableSortIndicesInt64Narrow/1048576/0/1/4', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 24, 'chunks': 4.0, 'columns': 1.0, 'null_percent': 0.0}
   TableSortIndicesInt64Narrow/1048576/4/1/1  87.635M items/sec 20.321M items/sec   -76.811   {'family_index': 12, 'per_family_instance_index': 34, 'run_name': 'TableSortIndicesInt64Narrow/1048576/4/1/1', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 58, 'chunks': 1.0, 'columns': 1.0, 'null_percent': 25.0}
 TableSortIndicesInt64Narrow/1048576/100/1/1 151.551M items/sec 15.919M items/sec   -89.496  {'family_index': 12, 'per_family_instance_index': 33, 'run_name': 'TableSortIndicesInt64Narrow/1048576/100/1/1', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 99, 'chunks': 1.0, 'columns': 1.0, 'null_percent': 1.0}
   TableSortIndicesInt64Narrow/1048576/0/1/1 177.335M items/sec 16.015M items/sec   -90.969   {'family_index': 12, 'per_family_instance_index': 35, 'run_name': 'TableSortIndicesInt64Narrow/1048576/0/1/1', 'repetitions': 1, 'repetition_index': 0, 'threads': 1, 'iterations': 117, 'chunks': 1.0, 'columns': 1.0, 'null_percent': 0.0}

Component(s)

C++

Activity

  1. pitrou commented on Jun 20, 2023

    @pitrou
    MemberAuthor

    @benibus Can you take a look? I expect the cause to be quite simple actually.

  2. changed the title [-][C++] Large regression in sort performance[/-] [+][C++] Large regression in single-key table sort performance[/+] on Jun 20, 2023
  3. added this to the 13.0.0 milestone on Jun 20, 2023
  4. benibus commented on Jun 20, 2023

    @benibus
    Contributor

    Yeah... the single key path for tables (but not batches) was removed entirely - so that makes sense. I'll re-add it.

  5. added a commit that references this issue on Jun 20, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions