Norbert Luksa has posted comments on this change. ( http://gerrit.cloudera.org:8080/15051 )
Change subject: IMPALA-9226: Improve string allocations of the ORC scanner ...................................................................... Patch Set 3: (2 comments) http://gerrit.cloudera.org:8080/#/c/15051/3//COMMIT_MSG Commit Message: http://gerrit.cloudera.org:8080/#/c/15051/3//COMMIT_MSG@24 PS3, Line 24: Some results: > As discussed offline: These measurement were run with a debug build. Let' s Right, unfortunately when I run the queries in release build, the results are not so bright. There is still a little speedup, but it's far away from the speedup in the debug build. Will upload the exact numbers when I tested it with more runs. http://gerrit.cloudera.org:8080/#/c/15051/3/be/src/exec/orc-column-readers.h File be/src/exec/orc-column-readers.h: http://gerrit.cloudera.org:8080/#/c/15051/3/be/src/exec/orc-column-readers.h@198 PS3, Line 198: blob_ = nullptr > So we cannot reuse blobs? Looking at the ORC library, it should be possible, since it passes the dictionary along the way. However, I did not find a simple solution for the same in Impala. After some digging, I found that the memory where the blob is currently allocated is owned by the row batch, and we cannot use this memory with the next row batch. I'll try to check if I keep track of the row batch and reuse the blob between them, will come back to you with the results. -- To view, visit http://gerrit.cloudera.org:8080/15051 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: If2d975946fb6f4104d8dc98895285b3a0c6bef7f Gerrit-Change-Number: 15051 Gerrit-PatchSet: 3 Gerrit-Owner: Norbert Luksa <norbert.lu...@cloudera.com> Gerrit-Reviewer: Csaba Ringhofer <csringho...@cloudera.com> Gerrit-Reviewer: Daniel Becker <daniel.bec...@cloudera.com> Gerrit-Reviewer: Gabor Kaszab <gaborkas...@cloudera.com> Gerrit-Reviewer: Impala Public Jenkins <impala-public-jenk...@cloudera.com> Gerrit-Reviewer: Norbert Luksa <norbert.lu...@cloudera.com> Gerrit-Reviewer: Zoltan Borok-Nagy <borokna...@cloudera.com> Gerrit-Comment-Date: Fri, 24 Jan 2020 15:51:05 +0000 Gerrit-HasComments: Yes
