There are some considerations the user should pay attention, in order to take full advantage of this tool. When performing a data pump export operation it can perform faster if using parallelism, but if this is not used properly, the process may end up serializing, which is very likely to happen if the dump files are written to the same disk location.
When performing a data pump import operation, we should consider the same parallelism issue. If using an enterprise edition, the degree of parallelism can be set and can be tuned so that there will be several parallel processes carrying out the import process. It is advisable to ensure the number of processes does not exceed twice the number of available CPU's.
Also, the tablespace features are important. The tablespace should be locally managed with Automatic Segment Space Management(ASSM); this will allow the insert process to perform faster.
Other features that should be considered are related to database block checking. Both db_block_ckecking
and db_block_checksum
impose a performance penalty. It has been reported by some users that this penalty is meaningful when batch loading takes place. It is advisable to either disable these parameters or reduce the emphasis. Those instance parameters are dynamic, so they can be modified during the operation.
Other instance parameters to consider are those related to parallelism, the parallel_max_servers
, and parallel_execution_message_size
. When using parallelism, the large_pool_size
region should be properly configured.