• Inter-node (pipelined) parallelism – an exchange node that spawns a single thread effectively pipelines rows from the plan fragment below the node to the plan fragment above the node. For example, an exchange node below a sort node allows the plan fragment below to generate rows at the same time as sorting is being done for previous rows. Plan fragments that produce and consume rows at the same rate can effectively overlap their processing, reducing the overall execution time for the query.

• Inter-node (bushy) parallelism – exchange nodes inserted over essentially independent query plan fragments allow those fragments to execute independently of one another. A specialized case of bushy parallelism occurs in union queries when a single exchange node is placed above the unioned selects. One thread is created for each of the select plan fragments, allowing the selects to be processed concurrently.

• Intra-node (partitioned table) parallelism – a single exchange node is placed above the orig node for a partitioned table. The exchange node creates several (4, 8, and so forth) child threads, each one of which retrieves data from a subset of the partitions of the table. This allows the concurrent reading of rows from the different partitions, clearly reducing the elapsed time required to process the table. A variation on partitioned parallelism (called a partition compatible join) occurs when two partitioned tables with the same number of partitions are joined using their respective partitioning columns. The query optimizer places the exchange node above the join in the query plan, resulting in mini-joins being performed between the rows of compatible pairs of partitions. These mini-joins are performed concurrently on different threads.