You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We use the Arrow IPC data format for shuffle operation in our project https://github.com/oap-project/gluten. To evaluate the performance, we run TPCH benchmarks of SF2T on a bare metal machine. By immediately shrinking the buffer after buffer compression, we observed that the average memory usage decreases 31%, and the total runtime decreases 7%.
Component(s)
C++
The text was updated successfully, but these errors were encountered:
…er and shrink after compression (#36294)
### Rationale for this change
Described in issue #36293#34025.
### What changes are included in this PR?
* Allocate buffer for compressed data using the memory pool given by the user
* Shrink compressed data buffer after compression to conserve memory, as the compressed data might be much smaller than the theoretical max compressed data size
### Are these changes tested?
Covered by existing tests.
### Are there any user-facing changes?
No.
* Closes: #36293
* Closes: #34025
Authored-by: Rong Ma <rong.ma@intel.com>
Signed-off-by: Antoine Pitrou <antoine@python.org>
Describe the enhancement requested
We use the Arrow IPC data format for shuffle operation in our project https://github.com/oap-project/gluten. To evaluate the performance, we run TPCH benchmarks of SF2T on a bare metal machine. By immediately shrinking the buffer after buffer compression, we observed that the average memory usage decreases 31%, and the total runtime decreases 7%.
Component(s)
C++
The text was updated successfully, but these errors were encountered: