When a query is issued against an Oracle Real Application Clusters (Oracle RAC) database with multiple nodes, the parallel processes may be spawned on different nodes. This approach is used to ensure that no one node becomes overloaded and that the processing power of all nodes is utilized as efficiently as possible.
However, under certain conditions, the interinstance traffic in the Oracle RAC database may already be significantly high. As the parallel processes on different nodes send their result sets via the interconnect, there is a strong possibility that this added traffic will introduce performance issues, especially related to global cache metrics. In such cases, you may want to restrict the parallel processes to the node where the parallel query coordinator runs. Because all the components of the query—the coordinator and the parallel processes—are in the same instance, there is no interinstance traffic and hence there are no global-cache-related issues.
The parallel_force_local parameter restricts parallel processes to a single instance. The default value is FALSE, meaning that the parallel processes can go into any available instance. To restrict the parallel processes to a single instance, set the parallel_force_local parameter value to TRUE
-Geek DBA
Follow Me!!!