Parallel Query Tuning - Expert Oracle RAC Performance Diagnostics and Tuning - page 243

■ there are several new parameters introduced in Oracle Database 11g release 2; we discuss these parameters
later in the appropriate sections. a few of the new parameters are listed following.
NAME TYPE VALUE
------------------------------------ ----------- ----------------
fast_start_parallel_rollback string LOW
parallel_degree_limit string CPU
parallel_degree_policy string AUTO
parallel_force_local boolean FALSE
parallel_io_cap_enabled boolean FALSE
parallel_min_time_threshold string AUTO
parallel_servers_target integer 256
recovery_parallelism integer 2

Note

Parallelization is dependent on the current data conditions such as volume and distribution, including data
partitions, indexes, resources, number of instances, and so forth. Consequently, when data changes, if a more optimal
execution plan or parallelization plan becomes available, Oracle will automatically adapt to the new situation.
Passing of data back and forth between the various processes is done using table queues (TQ). TQ is an abstract
communication mechanism that allows child data flow operations to send rows to its parents. Once the QC receives
the results back, it passes them over to the user that made the original request.
Figure 8-4 represents basic PQ architecture. The slave set processes (P0 and P1) read data from disk, and, using
the queue references and process queues, pass it to P3 and P2 for a sort and merge operation, if the query has a
GROUP BY or ORDER BY clause. Once this operation is complete, slave sets pass through the queue reference layer TQ
one last time before returning data back to the QC process. The QC then presents the results back to the client.