frecover performance

[Sebastian Haglund]

I have a T500 server running HP-UX 10.20 with an external DDS2 drive. fbackuping /home (1.06 Gb, 30085 files and directories) takes 23 minutes but frecovering everything back to an empty directory takes over 1 hour.
Is there a way to make the recover go faster or does it really take 3 times longer to restore than backup?
Since a full backup of the system takes 2 hours a restore would take about 6, I really need it to go faster.

[Stefan Farrelly]

Remember backups are normally designed to write quickly, not restore quickly - working on the premise you want a log of the former and not too much of the latter.

For frecover performance is depends on the frequency of checkpoints on the tape - which allows the tape to 'jump' when looking for stuff to restore/frecover. We use the following config file for fbackup which seems to offer the best of backup times and recover times;

blocksperrecord 128
records 32
checkpointfreq 256
readerprocesses 4
maxretries 1
retrylimit 0
maxvoluses 1000

Check the manpage for these, you can adjust even more for faster recoveries if you wish but at the cost of backup speed.

[A. Clay Stephenson]

This is rather typical of restores, especially those with have a large number of small files. There is a fairly large overhead associated with updating directories during the restore that the backup does not have to do. You would see much the same result using a high performance backup system like OmniBack II. Another difference is that the backup has multiple reader processes.

[Steve Steel]

Hi

Tune it and it should speed up

+ Create a suitable configuration file called config in the
directory /var/adm/fbackupfiles call with -c in fbackup.

Options
-c config config is the name of the configuration file, and can
contain values for the following parameters:

o Number of 1024-byte blocks per record.
o Number of records of shared memory to allocate.
o Number of records between checkpoints. Since the
EOF marks between checkpoints are also used for
fast searching on DLT-format drives, changing the
checkpoint frequency may also affect selective
recovery speed (see WARNINGS section).
o Number of file-reader processes.
o Maximum number of times fbackup is to retry an
active file.
o Maximum number of bytes of media to use while
retrying the backup of an active file.
o Maximum number of times a magnetic tape volume
can be used.
o Name of a file to be executed when a volume
change occurs. This file must exist and be
executable.
o Name of a file to be executed when a fatal error
occurs. This file must exist and be executable.
o The number of files between the fast search marks
on DDS-format tapes. The cost of these marks are
negligible in terms of space on the DDS-format
tape. Not all DDS-format devices support fast
search marks.

Each entry in the configuration file consists of one
line of text in the following format: identifier, white
space, argument. In the following sample configuration
file, the number of blocks per record is set to 16; the
number of shared memory records is set to 16; the
checkpoint frequency is set to 256; the number of file
reader processes is set to 2; the maximum number of
retries of an active file is set to 5; the maximum
retry space for active files is set to 5,000,000 bytes;
the maximum number of times a magnetic tape volume can
be used is set to 100; the file to be executed at
volume change time is /var/adm/fbackupfiles/chgvol; the
file to be executed when a fatal error occurs is
/var/adm/fbackupfiles/error; and the number of files
between fast search marks on DDS-format drives is set
to 200.

blocksperrecord 16
records 16
checkpointfreq 256
readerprocesses 2 (maximum of 6)
maxretries 5
retrylimit 5000000
maxvoluses 100
chgvol /var/adm/fbackupfiles/chgvol
error /var/adm/fbackupfiles/error
filesperfsm 200

Each value listed is also the default value, except
chgvol and error, which default to null values.


steve Steel

[Frank Slootweg · Accepted solution]

Adding to the other responses:

I think that things go about as fast as they can:

DDS-2 has a Native Mode (uncompressed) *maximum* speed of 500 KB/sec, i.e. 0.66 GB in 23 minutes. You get 1.06GB in 23 minutes, so compression is working and you get a speed of 805 KB/sec. I.e. you are apparently already using a suitable blocksperrecord setting.

As others have mentioned, it is the directory operations, especially if you have many files per directory, which is a large factor during restores.