[comp.os.vms] BACKUPs : the issues : speed/error checks/reassurance.

MACALLSTR@vax1.physics.oxford.ac.UK (07/18/87)

           BACKUP - speed, security, suggestions.
           -------------------------------------

The following are a few general comments on backing up disks.

Any back-up scheme is usually a TRADE-OFF between convenience and security and
 also between paranoia and reality.
Some commercial systems must take a complete backup at a given moment in time
 to maintain the INTEGRITY of the data and to ensure that all files are at
 precisely the same 'update' level - extremely important for financial systems,
 etc. Only a 'stand-alone' backup can meet this requirement as all activity must
 be frozen while the data is being backed up. There is no alternative to a break
 in service in such cases - may be conveniently scheduled outside working hours.
Many people like to have the REASSURANCE of the 'verify pass' to confirm that
 the data on disk and tape are the same. A 'verify pass' cannot make sense with
 the system active as files will inevitably change between being backed up
 and the running of the verify pass : the disk must be backed up standalone if
 a sensible verification is to be performed.
It may be reassuring to have the software perform some extra ERROR CHECKS. This
 may be done using /CRC in backup and also by having the /GROUP switch on. These
 are normal defaults with the BACKUP utility. The user assumes that, if there
 are tape or tape-drive errors, the software will make corrections. The 
 need for /CRC and /GROUP will depend on the backup medium being used : 1600bpi,
 tape, 6250 bpi tape, disk, etc.
Having MULTIPLE COPIES of each back-up tape may be considered essential for 
 some applications.
In other applications it is important to have the SYSTEM AVAILABLE while the
 backups are being run.
The SPEED of the back-up is probably of great importance in all cases whether or
 not the system is continuously available. A fast backup reduces the down-time
 if stand-alone is required and reduces the time tape drives are occupied if
 running while the system is 'live'.

To speed up BACKUP, the following are a few suggestions.
 
 (1) /BUFFER_COUNT=5 .

 (2) /NOCRC .  Not on 1600 bpi tapes.

 (3) /BLOCKSIZE=20480 . Large block-size also reduces number of tape mounts.
                        Beware of overrunnning end of tape on 1600 bpi tapes.
                        I've used this blocksize with 6250 for years without any
                        EOT mishaps.

 (4) MFA0:saveset,MFA1.  Use multiple drives for output. Saves about 5 minutes
            waiting for rewind on each tape. Mounting of one tape can be done
            while other is being written.

 (5) Make disk files contiguous. Use one of the many disk defragmentation
    packages to make all files contiguous. Saves some search time.

 (6) /NORECORD. If you use BACKUP/SINCE=datetime ( not /SINCE=BACKUP ) there
    is no real need for the /RECORD qualifier and you can make a huge impact
    on backup time with this. Our RP07, which has 30K+ files on it takes about
    one hour just to do the BACKUP date recording!

 (7) /NOGROUP. I wouldn't recommend this for tapes! It might be useful if
   you're using disk as the backup medium.

 (8) /NOVERIFY. Saves a vast amount of time.

 (9) /NOJOURNAL. The writing of a journal file is an overhead. However, some
   record of the names of the files you've got on your back-up tapes is
   probably essential for most people.

 (10) All the above suggestions are aimed mainly at FULL backups and assume 
    that one is using /IMAGE.

 (11) /PHYSICAL. This may be useful for standalone backups but beware of the
   restrictions on restoring from this type of backup.


Final comments.

 How real are worries about 'special' files not being correctly
   backed up in a 'live' backup?

 How real are worries about relying on the tape-drive's error correction
   mechanisms?

 How real are worries about the reliability of the tape itself, assuming that
   a good quality tape is being used?


 If you know that the probability of failure resulting in loss of data is no 
   more than that of a Jumbo Jet's engine falling on your head, I'd recommend
   that you at least try some of the above time-saving techniques. There are
   many more worth-while activities in life than watching tapes spinning for
   hours on end - even if you're being paid for it!

John