Maximizing Export Performance

Many DBAs are faced with the challenge of speeding up utility functions such as export. Typically, an organization has only a small window for maintenance, and utility jobs must complete within that timeframe. Fortunately, there are a few things a DBA can do to expedite exports. These include:

• Use Direct Path – Direct path exports (DIRECT=Y) allow the export utility to skip the SQL evaluation buffer, whereas the conventional path export executes SQL SELECT statements. With direct path, the data is read from disk into the buffer cache, returning rows directly to the export client. This can offer substantial performance gains, depending on the actual data. When using the direct path, the recordlength parameter should also be used to optimize performance.

• Use Subsets – By subsetting the data using the QUERY option, the export process is only executed against the data that needs to be exported. If tables have old rows that are never updated, the old data should be exported once, and from that point only the newer data subsets should be exported. Subsets cannot be specified with direct path exports since SQL is necessary to create the subset.

• Use a Larger Buffer – For conventional path exports, a larger buffer will increase the number of rows that are processed between each physical write to the export file. Fewer physical writes equals greater performance. The following formula can be used to determine a proper buffer size:

buffer size = rows in array * max row size

• Separate Tables – Separate those tables that require consistent=y from those that don’t, in order to expedite the export. This way, the performance penalty will only be incurred for those tables that actually require it.

For the table with one million rows, the following benchmark tests were performed using the different export options.


Table 4.1- Benchmark tests performed using the different export options.

The table above reveals a small improvement in performance was obtained by increasing the buffer size on a conventional export. Using direct=y offered no performance boost over conventional, until it was accompanied by recordlength, which reduced the elapsed time by 25 percent.

Once data has been successfully copied to an export file, it can then be used by the import utility, as described in the next section.

SOURCE : http://www.remote-dba.net/teas_rem_util11.htm