Details
-
Task
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
In order to avoid memory overuse ahead of time, which may lead to executor failure, we have to safely separate the portions of memory that the JVM heap, off-heap and 'slack' share. The current implementation ([PR: NEMO-388|https://github.com/apache/incubator-nemo/pull/223]) does not guarantee this, allocating the off-heap memory on-demand. Off-heap region size will be given as a ratio of executor memory just like the slack size. Since off-heap memory is used in various types of Blocks, off-heap memory should be retained.