Github user kiszk commented on the issue:
https://github.com/apache/spark/pull/19222
As @hvanhovell pointed out, the first implementation introduced a lot
polymorphic call sites in very performance critical code (e.g.
`getBaseObject()` or `getBaseOffset()`. While `MemoryBlock` class has
subclasses, `getBaseObject()` and `getBaseOffset()` have a single
implementation in `MemoryBlock` without polymorphic call site. Thus, I will
take option 1 by using a micro benchmark program :)
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
