Don't know whether this would work or not but if this involves some very
large files that get backed up every night,  would it be possible for
Daniel to use the ADSM archive function against these files, running a
couple of archives concurrently against files in the same filespace ?
If this worked, these files could be excluded from the daily incremental
backup process.  Anybody know whether this would work or not ?

Tim Pittson


>----------
>From:  Paul L. Bradshaw[SMTP:plbrads AT IBM DOT NET]
>Reply To:      ADSM: Dist Stor Manager
>Sent:  Tuesday, September 16, 1997 2:24 PM
>To:    ADSM-L AT VM.MARIST DOT EDU
>Subject:       Re: Shortening a long backup
>
>> From: Daniel Thompson <thompsod AT USAA DOT COM>
>
>> I intend to do some testing to find alternatives.  One question I have is
>> simply whether running 2 backups at the same time for the same node with
>> the same opt file would run significantly faster than just one?
>
>This will not work correctly.  If you have two DIFFERENT filespaces (drives
>or filesystems)  you can run incremental backups in parallel for the two,
>but running incremental backups simultaneously on the SAME filespace will
>cause many files to be backed up twice.  The reason is ADSM does a query to
>the server at the beginning of the backup to get the last state of the
>system, and then uses this to determine what files to backup.  If you run
>two backups in parallel the query will return almost the same info
>depending on the timing, thus causing files to be backed up twice.  This
>state is constantly updated during the backup process so waiting until the
>first one is half done will not help either.
>
>If you have a system with a few large files or a fairly static directory
>structure, you can run multiple selective backups in parallel, and this
>will work fine and give you a performance boost depending on CPU/network
>power.  Just specify the files to backup on the command line using
>wildcards.  You may wish to run an incremental backup after the selective
>to catch any files you may have missed.
>
>Paul L. Bradshaw
>PLBrads AT ibm DOT net
>