Hi @Sambavi , You can install any required dependencies on all nodes and use them but you need to keep in mind that Pandas and Numpy doesn't provide distributed computing option and it wouldn't work with big data sets. If your zeppelin configured to use yarn cluster mode It will take all data to spark driver in data node where spark driver located and try to process it there. (if its not big data set you can increase driver resources and it will work but its not looks like solution) if you use client mode it will take everything in zeppelin node.   I recommend to try HandySpark https://github.com/dvgodoy/handyspark  ... View more

It depends what you want to change: If you want just to add additional disks in all nodes follow this:    Best way to create partitions like /grid/0/hadoop/hdfs/data - /grid/10/hadoop/hdfs/data and mount them to new formatted disks (its one of recommendation parameters for hdfs data mounts but you can change it): /dev/sda1 /grid/0 ext4 inode_readahead_blks=128,commit=30,data=writeback,noatime,nodiratime,nodev,nobarrier 0 0 /dev/sdb1 /grid/1 ext4 inode_readahead_blks=128,commit=30,data=writeback,noatime,nodiratime,nodev,nobarrier 0 0 /dev/sdc1 /grid/2 ext4 inode_readahead_blks=128,commit=30,data=writeback,noatime,nodiratime,nodev,nobarrier 0 0 After that just add all partitions paths in hdfs configs like: /grid/0/hadoop/hdfs/data,/grid/1/hadoop/hdfs/data,/grid/2/hadoop/hdfs/data But dont delete existed partition from configuration because you will lost data from block which stored in   /hadoop/hdfs/data. Path dont really matter just keep them separately and dont forget to make re-balance between disks. ... View more

Best way is to join your nodes using SSSD service it will solve users directory creation problem + group mapping.  ... View more