On Fri, 10 May 1996 18:10:34 EDT "Steven P. Roder"
<[log in to unmask]> said:
>The problem here is that VM:backup detects changes to the data as it is
>being dumped, so it rewinds the tapes, and even remounts tapes, if
>needed, and trys again, for a configurable number of times, after which
>the disk or filespace is either skipped, or dumped physically,
This is seriously brain dead. Since tapes are at best almost as fast as
the disks, your chances of succeeding in making that kind of backup on a
large active minidisk are close to zero. With the 500-1500 cyl disks that
were mentioned, this algorithm is just a sophisticated way to do CP SLEEP
180 MIN, and then make a physical backup, which does *not* have better
integrity. Not only do you waste a lot of time, but the resulting backup
has sub-optimal integrity, compared to what could have been achieved with
file-level retries.
>Our filestores are in SFS, which exacerbates the problem, since the
>dumps are a bit slower.
But with SFS the file wouldn't change during the backup (that is, SFS
will shadow the file so that the copy you see retains integrity). Of
course, if VMBACKUP queries SFS explicitly for new changes, that's
another story. Also, on a notebook SFS directory, only a handful of files
will be changing during the backup operation.
Anyway, this is a completely idiotic algorithm. What good is a DDR-like
backup of a very active minidisk? The files that don't change will be ok
(unless the directory data is corrupt in the dump, which *can* happen,
although it is very unlikely), and the files that change will give you an
error 3 reading file or will contain garbage, and the only way to know
which files were affected is to review them manually. A file level backup
with file level retries (which could be grouped for small files, for
better performance) could have at least identified the corrupt files for
your verification.
Eric
|