Hi all,

This is not a new issue, I have mentioned it in issue[1] and [2].

Currently, in the internal IoTDB engine, we use String to refer to a time
series path.
Then, we have to either:

- save the substring of a time series path string: the storage group name,
the device name, the measurement name and the full path;
- save the full path only and split the string if needed, (vice-visa, sate
the sg name, device name, and measurement name separately and contact them
if needed).

Another concern is that we have to repeat to save these strings in the
TsFiles.. and then enlarge the tsfile size.

Therefore, I'd like to use an ID ( a long data type) to refer to a time
series. When IoTDB receives a SQL, it converts the string of time series
path to Long, and when IoTDB displays the result, it converts the ID back.

The drawback (as well as an advantage) is, we need to store the ID into
TsFiles, rather than the String path (i.e., root.xxx.xxx.xxx.xxx). So, if
someone wants to use the TsFile directly to Spark or Hive, then he/she has
to attach a dictionary file..

This is what I want to discuss with you, is this acceptable for users?

(By the way, I know Yanzhe had tried to implement this feature but failed
because it reduces the write performance a lot. But I still want to try
again... to control the side-effect).

How do you think?


[1] https://issues.apache.org/jira/browse/IOTDB-85
[2] https://issues.apache.org/jira/browse/IOTDB-164

Best,
-----------------------------------
Xiangdong Huang
School of Software, Tsinghua University

 黄向东
清华大学 软件学院

Reply via email to