Restore of a single file is too slow using Direct Access Recovery on a NetApp filer

Article: 100005554
Last Published: 2012-04-06
Ratings: 0 0
Product(s): Backup Exec

Problem

Using the Enable Direct Access Recovery option when restoring data from a large backup set from a NetApp filer slows down the restore

Cause

Without DAR (Direct Access Recovery), a restore operation of a NetApp must read the entire data set from the backup media. The NDMP agent on the NetApp server then only restores requested the portions of the backup while it is processing the entire data stream.  The purpose of the DAR feature is to change this functionality such that the restore operation quickly seeks directly to the position on the backup media where the requested data resides, and then only reads that data to complete the restore operation.
 
The DAR restore should be much faster if only a small portion of the backup set is being restored because only that data has to be read.  In practice the read portion of the process when using DAR is extremely slow when coming from tape.  The NetApp NDMP server is reading relatively small chunks of data from the tape (1MB).  In between each request for data, the tape underruns and has to reposition.  This results in a shoe-shining action on the tape which greatly impacts the overall performance of the reads.
 
When DAR is turned off for the restore operation, the restore has to read all of the media to complete the restore operation. One of the side effects of this is that, if the data to be restored is near the front of the data set then the file being restored may be available long before the restore operation itself finishes reading the rest of the media. However, non-DAR restore operations will outperform DAR restore operations when larger data sets are being restored.
 

Solution

WORKAROUND:
In general, it is best to separate the backup into smaller pieces.  This improves the restore times.
Then, during the restore operation, a decision needs to be made on whether or not to use DAR on the restore operation.  This is set in the NDMP tab of the Restore Job Properties dialog.  If the amount of data to be recovered is small (a few files), then using DAR will be helpful due to there is little data to read. If the amount of data is large (many directories, or even the entire backup set), turning off DAR is recommended due to using DAR will impact on the performance of the restore operation. Note that in this case, the entire backup set will be read during the restore operation. This is why having a smaller backup set is recommended above.
Another alternative is to use disk media as the target for the backup.  Restoring from disk doesn't have the problem of underrun/reposition.  Although reading data with the DAR option enabled is still slower than without it.

Was this content helpful?