Each partitioned range has a dedicated consumer process that sorts the data in that range independently of other ranges. Depending on the size of the table and the number of buffers available to perform the sort, the consumers may perform multiple merge runs, writing intermediate results to disk, and reading and merging those results, until all of the data for the assigned range is completely sorted.
For create index commands, each consumer for each partitioned range of data writes to a separate database device. This improves performance through increased I/O parallelism, if database devices reside on separate physical devices and controllers. The consumer process also builds an index, referred to as a subindex, on the sorted data.
For merge joins, each consumer process writes the ordered rows to a separate set of linked data pages, one for each worker process that will perform the merge.
For queries, the consumer process simply orders the data in the range from the smallest value to the largest.
