HP 12000 Design Manual page 90

d. the type of backup – full or incremental
e. the type of data in the backup – files, database, etc.
The deduplication software then queries the metadata database to find an equivalent older
version of the same backup job to compare it against the new backup. If the current backup
is full, it will be compared against the last equivalent full backup version. If the current backup
is differential or incremental, on systems running firmware version 3.3 or higher it will be
compared against the previous incremental or differential. (TSM "incremental forever" backups
are all treated as full backups by deduplication.)
3. Data comparison
This phase is also called delta differencing. After the data grooming phase is complete, and
if there is an older version of the backup, either full or incremental, the new backup is
compared, or delta-differenced against it. The deduplication software looks for differences at
the byte level between backup objects with two different differencing schemes depending on
the type of the backup:
File-level differencing is for standard file backups and compares any changed file against
the previous version of that file at the byte level. This improves deduplication performance
because it does not have to compare unchanged files.
Backup-level differencing is for any non-file backup types (all database agent backups
and any unknown agent backups) and compares across the entire backup at the byte
level. This differencing method can optionally be used for file backups that contain
constantly changing file names, such as database dump files, because the differencing
is done at the job level without any awareness of file names within the backup.
4. Data reassembly
In this phase, duplicate data is replaced with pointers, and pointers are readjusted so that
they point to the most recent instance of data.
If the new backup is a full backup, this phase does not touch the new data, so that the most
recent backup is complete and available for restores. Instead, a new copy of the older version
is created. See
If the new backup is incremental, it is reassembled so that duplicate data is replaced with
pointers to the last full backup version.
90 Accelerated Deduplication
Figure 43 (page 91).