Yuan Tian created IOTDB-4005:
--------------------------------

             Summary: Parallel the FragmentInstance execution
                 Key: IOTDB-4005
                 URL: https://issues.apache.org/jira/browse/IOTDB-4005
             Project: Apache IoTDB
          Issue Type: Improvement
            Reporter: Yuan Tian
            Assignee: Yuan Tian


Previously, we accelerate raw query data without value filter by split each 
sensor's scan execution into a thread pool, in this way, we can accelerate the 
decompress and decode phase in query processing.

In current mpp, we only run one fragmentInstance in one thread, so all sensors 
in that FragmentInstance will only scan, decompress and decode one by one which 
cause the query performance won't reach the previous level.

We need to rethink about how to parallel our FragmentInstance, and meanwhile we 
should also take memory control into account, the previous version may cause 
OOM problem while there exists too many sensors in one query.

We should also take deadlock into account while doing memory controlling, in 
case of one sensor occupying all the remaining usable memory, another sensor 
can never get more memory, but the whole query must wait for each sensor 
returning at least one TsBlock.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to