In a Hadoop cluster, should Hive LLAP daemons work on datanodes or on dedicated nodes?

Question

In a Hadoop cluster, should Hive LLAP daemons work on datanodes or on dedicated nodes?(as the only service on node) Impala daemons are suggested to be installed on each datanode, however this article suggests instaling LLAP daemons on dedicated nodes

Answer 1

LLAP instances have dedicated Hive daemons for faster access. While you could share resources between regular node managers and LLAP daemons you would get better performance if all of the node is dedicated to it. Note that you still need to run node manager on the node because LLAP Daemons are containers and not actual OS processes. You can also run data node on the machine if you have storage on the node that you'd like to use.

TL;DR - Run on a few dedicated nodes for better performance