In running a test restoral of an entire AIX 3.2 file system to empty space I
watched the whole event in both ADSM and AIX to see what was happening.  Both
incremental backups and the restoral were done with the default management
class, and it is our practice for backups to go directly to tape (rather than
to use an intermediary disk pool such that the tape contains everything needed
for a restoral, with no data possibly lost if anything happens to the disk
pool).  One sees the restoral occurring in three phases:
  1. The directory structure gets recreated, very early in the task.
  2. There is a long period (approximately half of the overall task) in
     which dsmserv processes are busy, but no data is being restored.
     Presumably, ADSM is searching its database to round up all the file names
     belonging to the filespace, and getting them into the same order that the
     files appear on the tape.
  3. Data restoral actually occurs.  The 3590 tape on which the collocated
     data resided was mounted only at this point.

What I found particularly interesting was that the directories were recreated
without mounting or accessing the data tape.  This leads me to the conclusion
that basic info about directories is stored in the database and thus the
directories can be immediately redefined just from the database...which makes
perfect sense in that a directory is just another database object like a file,
except that a directory contains no data itself until all its files return,
and thus all info about the directory can totally reside in the database.
Could someone in IBM say whether or not this is the case?

If this is the case, then it's not clear to me just what effect the mysterious
DIRMc client option has.  There is scant information about just what effect
DIRMc produces, but seems to suggest that its purpose is to send directory
information to a certain backup storage pool.  But, again, if directory
information is kept in the database, then I don't see that the directory info
would also be sent to a storage pool - it would seem pointless.  Can anyone
clear this up?

    thanks,  Richard Sims, Boston University OIT