[
https://issues.apache.org/jira/browse/IOTDB-3884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17572940#comment-17572940
]
Quan Siyi commented on IOTDB-3884:
--
When executing cross sapce compaction and inner space compaction, we need to
decode and decompress chunk of timesereis in the compaction task. We use
sereisReader to get chunk of tsfile and decompress and decode it, and return
BatchData for further process. Now We the new util "scanSeriesUtil" to get
TsBlock of timeseries. The only difference between of BatchData and TsBlock is
that Batchdata is row-based and TsBlock is column-based. Therefore, if we
continue processing data in a row-based format, we spend much time converting
tsBlock to rows.
> [Performance]Cross_space and unseq_space compaction are slower at aligned
> timeseries
>
>
> Key: IOTDB-3884
> URL: https://issues.apache.org/jira/browse/IOTDB-3884
> Project: Apache IoTDB
> Issue Type: Bug
> Components: Core/Compaction
>Affects Versions: 0.14.0-SNAPSHOT
>Reporter: FengQingxin
>Assignee: Quan Siyi
>Priority: Major
> Attachments: image-2022-07-19-14-17-03-521.png,
> image-2022-07-19-14-29-01-699.png, image-2022-07-19-14-30-03-267.png,
> screenshot-1.png
>
>
> Please refer to below picture:
> Test scenario:Build and Start an iotdb instance of each commit id in master
> branch with one dataset, statistics the compaction time in logs.
> !image-2022-07-19-14-30-03-267.png|width=928,height=362!
> !image-2022-07-19-14-29-01-699.png|width=931,height=187!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)