Details
-
New Feature
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
-
2021-12-Dragon
Description
Different with traditional relational databases, IoTDB uses tree-based structure in memory to manage the schema (a.k.a, metadata), and use a Write-Ahead-Log-like file structure to persist the schema.
Now, each time series will take about 300 Bytes in memory. However, an IoTDB instance may manage more than 100 million time series, which may take more than 30GB memory.
Therefore, we'd like to re-design the schema management module.
1. File: Persist the tree on disk like a b-tree.
2. Cache: we may have no memory to load the whole tree. So a cache is needed and query from the tree on disk is needed.