Optimize Oracle import (data pump impdp) performance Don Burleson

Note:  This is a quick overview of improving data pump import performance for beginners.  For expert-only tips on improving import speed, see here.  Also see my notes on tuning and improving Oracle export (expdp) performance speed.

Regardless of which options were used when the data was exported, it has no influence on how the data is imported.  For example, it is irrelevant to the import process whether it was a direct path export or not, since it is a plain export file, be it generated from direct or conventional means.  

Unfortunately, there is no direct option available for imports (only for export and SQL*loader).  The import process has more tuning limitations than other utilities.  The DBA should consider the following when trying to optimize import performance:

Set commit=n – For tables that can afford not to commit until the end of the load, this option provides a significant performance increase.  Larger tables may not be suitable for this option due to the required rollback/undo space.

Set indexes=n – Index creation can be postponed until after import completes, by specifying indexes=n.  If indexes for the target table already exist at the time of execution, import performs index maintenance when data is inserted into the table.  Setting indexes=n eliminates this maintenance overhead.

Use the buffer parameter – By using a larger buffer setting, import can do more work before disk access is performed. 

Hidden parameters - You can also use the hidden parameter _disable_logging = true to reduce redo, but beware that the resulting import will be unrecoverable.

Import Speed Benchmarks:

 Oracle guru Steve Callan notes that he has run parallel Data Pump import jobs to load 1.8 terabytes (1,800 gigabytes) in less than a day, an import load rate of over 75 gigabytes per hour.

“The target box was an AIX 5L using LPARS, pretty sure it was 32 CPU and 64GB RAM. There were several distinct schemas, so the data pump export/import was a matter of divide and conquer (i.e., stream several jobs/sessions). I think the largest chunk was around 900GB.

Statistics gathering took a while, but that time was separate from the actual import, and some indexes were skipped/re-built later.  

An old import version of this lasted right around three days, and the data pump version was about a third of that time.”  

As we see, the import speed is far greater using the newer Data Pump import utility (impdp) and imports run faster of faster servers.

For more details on Oracle utilities, see the book "Advanced Oracle Utilities" by Bert Scalzo, Andrew Kerber, Donald K. Burleson, and Steve Callan. You can buy it direct from the publisher for 30% off directly from Rampant TechPress.

Also see: