I am reading following article about 2. Cache-aware Computation

https://www.databricks.com/blog/2015/04/28/project-tungsten-bringing-spark-closer-to-bare-metal.html

It provides a picture as below showing that key-pointer pair structure can help for the cache aware computation/sorting

But,In the key-pointer pair, spark still needs to refer to the remote value object in the main memory which still leads to cache loss, so I don’t understand how cache aware works here

Could some one help explain, understanding this picture will help my understanding of spark code..

https://www.databricks.com/wp-content/uploads/2015/04/Screen-Shot-2015-04-27-at-6.12.51-PM.png