[IOTDB-596] [Distributed] Better cache for remote schema

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Component/s: Core/Cluster
• Assignee: Unassigned
• Reporter: WangChao
• Created: 16/Apr/20 09:46
• Updated: 16/Apr/20 09:46

Description

Each timeseries has its own sampling period. For the timeseries with short sampling period, it should exist in cache for a long time. The LRU algorithm is based on the access frequency to eliminate the timeseries.

 

Therefore, the following scenarios may exist:

 

A timeseries generates one data per second, and B timeseries generates one data every 10 seconds. Maybe the 10th second,  A will be eliminated from the cache.

 

Consider a better cache algorithm, for example:

 

Add sample period data to each time series. When the time sequence needs to be eliminated, compare the sampling period, and leave the short time.

 

As for how to obtain the sampling period, a simple point can sample the time interval between two accesses of the cache.

Issue Links

• is related to: IOTDB-586 [Distributed] Eliminate some timeseries schemas When memory is limited