Avamar: Optimising performance when backing up Outlook archive .pst files

This article discusses how backing up .pst files impacts on the performance of an Avamar backup. It also discusses ways to assess and potentially mitigate poor backup performance.

An Avamar client with many Outlook archive (.pst) files may back up more slowly than a client which contains more typical file system type data.

Below are examples of performance data from backups which contain a large proportion of .pst files.

Example 1:
2014-06-23 16:36:45 avtar Info <5156>: Backup #4 timestamp 2014-06-23 16:36:44, 2,654,547 files, 685,600 folders, 1,935 GB (46,371 files, 9.027 GB, 0.47% new)
2014-06-23 16:36:45 avtar Info <6083>: Backed-up 1,935 GB in 1176.67 minutes: 99 GB/hour (135,359 files/hour)


Example 2:
2014-06-19 08:38:56 avtar Info <5156>: Backup #1626 timestamp 2014-06-19 08:39:00, 4,279,104 files, 491,108 folders, 1,427 GB (206,579 files, 2.002 GB, 0.14% new)
2014-06-19 08:38:56 avtar Info <6083>: Backed-up 1,427 GB in 878.94 minutes: 97 GB/hour (292,110 files/hour)

Although the change rate in each example is modest (0.47% and 0.14%), the performance is 292,110 files per hour. This is well below the 1 million files per hour considered typical performance where data is read from rotational drives.

Customer configuration

To check the impact which .pst files have on backup performance, exclude them from the dataset and create a dataset exclusively for the .pst files. 

This allows you to use the avtar Info <5156> and avtar info <6083> client log entries to:

1) Learn the average (mean) size of the .pst files within the dataset.
2) Learn what proportion of the files are changing each day.
3) Learn how the backup performance of .pst files compares with that of other data.

The following techniques help mitigate slow performance of backups which contain .pst files.

Back up .pst files separately from general data:
Exclude .pst files from the regular dataset. The .pst file data is backed up in its own time and not interfere with the otherwise quick backup performance which is possible with regular files. In extreme scenarios, the .pst files could be backed up less frequently, perhaps only on weekends.

Use Image backup for virtual machine clients:
If the amount of data change within the backup is relatively low, consider backing up the client using an Avamar image proxy client. Image proxy clients that are configured with change block tracking monitor the changed blocks within the Virtual Machine's .vmdk file. Image backups do not scan every file and process those which were modified. This usually results in faster backup performance.

Back up email data directly from Exchange using the Avamar plug-in:
The preferred method to back up email data is to use the Avamar plug-in for Exchange. This results in more efficient backup performance and avoid the potential for backing up the same data more than once. If data in .pst files is already stored on an Exchange server which Avamar backs up, consider adding a global exclusion for .pst files.

When backing up .pst files through the Avamar file system plug-in, Microsoft VSS ensures that open .pst files are 'crash consistent*'. However, those files may not be 'application consistent'. Which is to say, the backup does not capture data in memory or capture pending I/O operations that are directed to the file.

Any files which are open during the backup process may not be 'consistent' when backed up.  

Backups of regular file system data on a typical Avamar client configured with rotational (non-SSD) drives typically perform at around 1 million files/hour. 

This value files/hour performance varies and depends on many factors including the nature of the client data, network throughput, disk-subsystem performance, and CPU availability.