As we discussed about the file and data management in HBase, along with compaction, Splitting Regions also is an important process. The best performance in HBase is achieved when the data is distributed evenly across the Regions and RegionServers which can be achieved by Splitting the Region optimally. When a table is first created with default options, only one Region is allocated to the table as HBase will not have sufficient information to allocate the appropriate number of Regions. We have three types of Splitting triggers which are Pre-Splitting, Auto Splitting, and Forced Splitting.
To aid the splitting of a Region while creating a table, we can use Pre-Splitting to let HBase know initially the number of Regions to allocate to a table. For Pre-Splitting we should know the distribution of the data and if we Pre-Split the Regions and we have a data skew, then the distribution will be non-uniform and can limit the cluster performance. We also have to calculate the...
