Disk-partitioning algorithm

If a simple NSD file is used without the -meta label, Ambari assigns metadata and data disks and partitions the disk according to the following rules:
  1. If node number is less than or equal to four:
    • If the disk number of each node is less than or equal to three, put all disks to system pool, and set usage = metadataanddata. Partitioning is not done.
    • If the disk number of each node is greater than or equal to four, assign metaonly and dataonly disks based on a 1:3 ratio on each node. The MAX metadisk number per node is four. Partitioning is done if all NodeManager nodes are also NSD nodes, and have the same number of NSD disks.
  2. If the node number is equal to or greater than five:
    • If the disk number of each node is less than or equal to two, put all disks to the system pool, and usage is metadataanddata. Partitioning is not done.
    • Set four nodes to metanodes where meta disks are located. Others are DataNodes.
    • Failure groups are created based on the failure group selection rule.
    • Assign meta disk and data disks to the meta node. Assign only data disk to the DataNode. The ratio follows best practice, and falls between 1:3 and 1:10.
    • If all GPFS nodes have the same number of NSD disks, create a local partition on data disks for Hadoop intermediate data.