In YARN, whenever a job is submitted, it sets up a distributed cache for jars and configuration files per job. What this means is that the jars will be cached during the execution life cycle of a job. However, often the jars or the code does not change across different users of the cluster.
To solve the problem of loading jars for every job, which consume network bandwidth, a proposal is in place to implement a shared cache across the cluster for all users to use it.
You will need a running cluster with HDFS and YARN set up properly so that the user can run test jobs such as pi or wordcount examples on it.
This feature is not yet production ready and is scheduled to be a standard feature in Hadoop 2.9.0, ...