Hi,
I am also confused about the "six times worse" when creating timeseries. Could you propose a draft pull request to IoTDB so that we can have a look at your code implementation? Sincerely, Lei Rui On 9/11/2019 09:25,安彦哲<thss15_a...@163.com> wrote: Hi, I'm trying to solve [IoTDB-164](https://issues.apache.org/jira/projects/IOTDB/issues/IOTDB-164?filter=allopenissues). To accomplish the task, I've modified the structure of MTree and replace Path(String) of several Maps with ID(Long), including latestTimeForEachDevice and latestFlushedTimeForEachDevice in the org.apache.iotdb.db.engine.storagegroup.StorageGroupProcessor.java memTableMap in the org.apache.iotdb.db.engine.memtable.AbstractMemTable.java Then, I've carried out a simple experiment: create 200,000 timeseries in the same storage group insert 10 tuples(<timestamp, value>) into each of timeseries The result shows that Step 1: the original version costs 25138.8ms while the modified one costs 177753ms. It concludes that the modified version performs six times worse than the original version when creating timeseries. Step 2: the original version costs 213662.8ms while the modified one costs 194271.2ms. It concludes that the modified version performs nine percent better than the original version when inserting values. In addition, we don't usually create too many timeseries, while insertion is the most frequently used operation in IoTDB. I'm confused whether we should apply the modified version. If anyone knows more about this topic, please inform me. Best, --------------------------------------------------- Yanzhe An School of Software, Tsinghua University | | 安彦哲 | | thss15_a...@163.com | 签名由网易邮箱大师定制