You have recently converted your Hadoop cluster from a MapReduce 1 (MRv1) architecture
to MapReduce 2 (MRv2) on YARN architecture. Your developers are accustomed to
specifying map and reduce tasks (resource allocation) tasks when they run jobs: A
developer wants to know how specify to reduce tasks when a specific job runs. Which
method should you tell that developers to implement?

A.
MapReduce version 2 (MRv2) on YARN abstracts resource allocation away from the idea
of “tasks” into memory and virtual cores, thus eliminating the need for a developer to specify
the number of reduce tasks, and indeed preventing the developer from specifying the
number of reduce tasks.

B.
In YARN, resource allocations is a function of megabytes of memory in multiples of
1024mb. Thus, they should specify the amount of memory resource they need by executing
–D mapreducereduces. memory-mb-2048

C.
In YARN, the ApplicationMaster is responsible for requesting the resource required for a
specific launch. Thus, executing –D yarn.applicationmaster.reduce.tasks=2 will specify that
the ApplicationMaster launch two task contains on the worker nodes.

D.
Developers specify reduce tasks in the exact same way for both MapReduce version 1

(MRv1) and MapReduce version 2 (MRv2) on YARN. Thus, executing –D
mapreduce.job.reduces-2 will specify reduce tasks.

E.
In YARN, resource allocation is function of virtual cores specified by the
ApplicationManager making requests to the NodeManager where a reduce task is handeled
by a single container (and thus a single virtual core). Thus, the developer needs to specify
the number of virtual cores to the NodeManager by executing –p
yarn.nodemanager.cpu-vcores=2