Interpartition parallelism with nickname serial fetching
Interpartition parallelism with serial fetching applies to the queries that reference local and remote data. The federated server can distribute the remote data to each of the local partitions.
Figure 1 and Figure 2 show the concept of interpartition parallelism that involves local and remote data sources.
Figure 1 shows how this type of query is processed without interpartition parallelism. The remote nickname data and the local partitioned data are processed serially at the coordinator partition. This technique does not exploit the parallel power of the database partitions because most processing is performed on a single partition. If data volumes are very large, this technique is likely to result in long-running queries.
Figure 2 shows how processing occurs when the optimizer distributes the nickname data to the partitions. The coordinator partition fetches the nickname data and distributes the data to the database partitions for parallel processing. When parallel processing is complete, the results are sent back to the coordinator partition for final processing before they are returned to the application.
Figure 3 and Figure 4 show the concept of interpartition parallelism that involves only remote data sources.
Figure 3 shows serial processing of the remote nickname data at the coordinator partition. The coordinator partition, which also acts as the federated server, retrieves the nickname data and process it serially.
Figure 4 shows the coordinator partition distributing the data across a computational partition group. Computational partition groups allow the optimizer to generate access plans that distribute nickname data across the partitions of a partitioned database server for processing in parallel.
