Skip to content

Performance degradation with repeated query execution #907

@quasiben

Description

@quasiben

An example of perf degrading over time (Q9)

___________________________________________________________________________________________________
  _ iter _ wall time _ spill D_H _ D_H time _ host alloc total _ host stream delay _ recv-into-host _
  ___________________________________________________________________________________________________
  _ 0    _ 68.4s     _ 491 GB    _ 238.7s   _ 796 GB           _ 43.4s             _ 122K           _
  ___________________________________________________________________________________________________
  _ 1    _ 74.4s     _ 528 GB    _ 259.9s   _ 876 GB           _ 47.9s             _ 133K           _
  ___________________________________________________________________________________________________
  _ 2    _ 84.6s     _ 517 GB    _ 275.4s   _ 995 GB           _ 49.9s             _ 172K           _
  ___________________________________________________________________________________________________
  _ 3    _ 96.5s     _ 536 GB    _ 277.8s   _ 1027 GB          _ 52.1s             _ 178K           _
  ___________________________________________________________________________________________________
  _ 4    _ 90.0s     _ 531 GB    _ 276.4s   _ 1019 GB          _ 56.7s             _ 174K           _
  ___________________________________________________________________________________________________

We are spilling and this is probably what's driving time up/perf down but why do we need to spill more over time ? We think we are correctly cleaning up between queries so we don't think it's a garbage collection issue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions