A tech startup has been custom building and configuring their Hadoop clusters for over a

decade. They have $100$T of data stored in HDFS$,$ with triple replication, that they run Hive

queries on$.$

With the recent economic downturn, the decision was made to move to AWS.

The lead engineer $($the one who has been most involved with keeping Hadoop running$)$

is pushing to allocate N EC$2$ nodes on AWS and configure HDFS and Hadoop on those:

essentially replicating the in$-$house environment, except using AWS hardware.

Another engineer is suggesting to store all data in S$3,$ and spin up EMR clusters whenever

there is a need to query data.

Questions:

$1.$ What are the pros$/$cons of either approach?

$2.$ Why did Hadoop couple compute and storage into the same system?

$3.$ Why is everyone decoupling compute from storage these days?