[ https://issues.apache.org/jira/browse/IOTDB-5265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17683737#comment-17683737 ]
周沛辰 commented on IOTDB-5265: ---------------------------- 复测并对整体进行抽样,发现Fast3没有数据删除耗时4690秒,Fast3有数据删除耗时4743秒,而Fast3有数据删除的测试中,合并调度的空闲时间耗时较久,扣除调度空闲时间,整体合并的耗时两者差不多。 !image-2023-02-03-16-13-25-462.png|width=633,height=331! > [fast compaction] One tenth of the devices are "ALL_DELETED",fast compaction > is slow > ------------------------------------------------------------------------------------ > > Key: IOTDB-5265 > URL: https://issues.apache.org/jira/browse/IOTDB-5265 > Project: Apache IoTDB > Issue Type: Improvement > Components: mpp-cluster > Reporter: 刘珍 > Assignee: 周沛辰 > Priority: Major > Attachments: 1.conf, 2.conf, image-2022-12-22-10-05-22-860.png, > image-2023-02-03-16-13-25-462.png > > > master 1219 4a59a2c > 100 设备 * 6 序列共600序列,2亿点/序列,顺序和乱序各50%。 > 对比2个场景的合并性能: > delete 10个设备的数据 > 没有delete操作 > 有如下2个问题, > 问题1: > cross_performer=fast配置:没有delete数据的场景,合并耗时最长,比默认都长。 > 问题2: > cross_performer=fast > inner_unseq_performer=fast > inner_seq_performer=fast > 配置 > 没有delete数据的场景,反而合并的快。 > !image-2022-12-22-10-05-22-860.png|width=942,height=232! > 测试环境: > 1. 192.168.10.72 48CPU384GB > 关合并, > time_partition_interval=60480000000 > 生成测试数据。 > 配置文件见附件,1.conf运行完,cli flush,再运行2.conf,cli flush。 > 2.对比合并性能 > compaction_write_throughput_mb_per_sec=1048576 (合并不限流) > time_partition_interval=60480000000 > query_timeout_threshold=36000000 > delete 10个设备的数据: > cli -e "delete from root.test.g_0.d_0.*;" > cli -e "delete from root.test.g_0.d_10.*;" > cli -e "delete from root.test.g_0.d_20.*;" > cli -e "delete from root.test.g_0.d_30.*;" > cli -e "delete from root.test.g_0.d_40.*;" > cli -e "delete from root.test.g_0.d_50.*;" > cli -e "delete from root.test.g_0.d_60.*;" > cli -e "delete from root.test.g_0.d_70.*;" > cli -e "delete from root.test.g_0.d_80.*;" > cli -e "delete from root.test.g_0.d_90.*;" -- This message was sent by Atlassian Jira (v8.20.10#820010)