1. Hadoop has a master-slave topology with one master node that assigns tasks to multiple slave nodes, which do the actual computing. The slave nodes store data while the master node stores metadata. 2. MapReduce is the processing layer that breaks jobs into independent tasks that can run in parallel on slave nodes. Map performs sorting and filtering of data while Reduce summarizes the output of Map. 3. YARN manages resources across clusters by allocating resources for applications through a resource manager and node managers that monitor resources on machines.