I'm using dbms_parallel_execute to insert records into a partitioned table (64 partitions).
* I created the chunks so that the chunks were created using "create_chunks_by_sql". The chunks were created so that the the first chunk is for the 1st partition, the 2nd chunk is for the 2nd partition, ... the 64th chunk is for partition 64, and this repeats. This helps to ensure the inserts are distributed across the partitions evenly.
* I incorrectly assumed that Oracle assigns the 1st chunk, the chunk with the lowest chunk_id number, and then the next and so on until completion. It appears that it selects a group of about 40 chunks in sequential order (unassigned) from anywhere (random) and then grabs another set and so on.
* I'm interested in running a performance test using a varying number of parallel processes to acquire the best number of processes. The test would be a little more accurate if the records sets (chunks) were processed in order.
1) What is the method Oracle uses for assigning the chunk for processing?
2) Is there a configuration setting to force oracle to do the chunks in a given order? I'm interested in doing as little programming to accomplish this. I know you can process the chunks using a 'user-defined framework".
3) Any other ideas?