Hi,
Can you be a bit more specific about MapReduce applications? Are these primary applications using Hive tables on prem. As I understand you get to build clusters with the chosen number of nodes for master + n workers. Then you have to migrate your HDFS storage (Hive tables or others) to buckets in Cloud (which is not plain sailing from my experience) and then use Spark on the storage or BigQuery tables. That will provide you with Ephemeral clusters of your choice to perform you work again predominantly Spark as the execution engine.
So it boils down on the usage of Spark as the execution engine on Prem already and you migration plan to Cloud. Also from my experience you will have to maintain a hybrid model for sometime, in other words your on-prem should co-exist with Cloud for sometime.
So it boils down what issues are you facing that you want to consider Dataproc as an option.
HTH,
Mich