[ https://issues.apache.org/jira/browse/IOTDB-4364?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17603471#comment-17603471 ]
周沛辰 commented on IOTDB-4364: ---------------------------- When compacting, it is necessary to obtain devices and timeseries from all source files, as well as the metadata of each timeseries. To avoid traversing the Zesong tree multiple times, we record the intermediate nodes of the Zesong tree. The comparison results are below: !image-2022-09-13-17-00-39-214.png|width=907,height=500! !image-2022-09-13-17-01-06-224.png|width=695,height=400! > Reduce read amplication in compaction > ------------------------------------- > > Key: IOTDB-4364 > URL: https://issues.apache.org/jira/browse/IOTDB-4364 > Project: Apache IoTDB > Issue Type: Bug > Reporter: 周沛辰 > Assignee: 周沛辰 > Priority: Major > Labels: pull-request-available > Attachments: image-2022-09-08-11-45-36-035.png, > image-2022-09-13-17-00-39-214.png, image-2022-09-13-17-01-06-224.png > > > In massive timeseries scenarios, each device has 20 timeseries, and each file > has tens of thousands of devices, that is, millions of timeseries. Read > throughput in cross compaction is 100 times greater than write throughput. > > !image-2022-09-08-11-45-36-035.png! -- This message was sent by Atlassian Jira (v8.20.10#820010)