Ordering with fragmented indexes

When an index is not fragmented, the database server can use the index to avoid a sort. However, when an index is fragmented, the ordering can be guaranteed only within the fragment, not between fragments.

Usually, the least expensive option for the total-query-time goal is to scan the fragments in parallel and then use the parallel sort to produce the proper ordering. However, this option does not favor the user-response-time goal.

Instead, if the user-response time is more important, the database server reads the index fragments in parallel and merges the data from all of the fragments. No additional sort is generally needed.