[OpenAFS] ADSM(TSM) Backup

Harald Barth haba@pdc.kth.se
Tue, 07 May 2002 11:57:45 +0200 (CEST)

I'm not reading openafs-info on a regular basis, but I got pointed
on this thread as we use TSM to backup AFS and have done so for 
some time.

> 3. The crucial point:I have to use ADSM(TSM) for backup. How can
>    this be done? XBSA? (we have no AIX client)

AFS backups can be done on 

* File level

  +per file restore
  +users can do restores themselves
  +file access through cache manager faster than volume access
  +you can restore to any kind of file system
  -you need to authenticate on file level
     one of
       process runnig as system:administrator
       process running as backupuser and you need to add "backupuser read"everywhere
  -running the client traversing volume mountpoints will not work as there is no
   loop detection in the client
  -running the client without traversing volume mountpoints: You need to assure that
   all your volumes are mounted somewhere
  -AFS functionality does not exist in all TSM clients platforms for which there
   was AFS available from the same company.

* Volume level

  -you need some kid of working AFS to restore the data
  -does not use any AIX specific software.

  With AFS backup and butc or buta
  +integrated with the rest of AFS
  -the AFS backup database is not a very clever one
  -there is a TSM backend on AIX, but very limited functionality

  With homegrown vos-dump-and-backup the dump
  +good control over what is backed up
  +separation between AFS and backup system, volume dumps can be
   examined between dump and backup
  -there is no IBM supported utility to backup anything from stdin - for
   example volumedumps - you have to write them to a temporary filesystem first
  -volume dump slower than file access

pdc.kth.se and e.kth.se _archive_ volume dumps into TSM through an
homegrown TSM-API application.

stacken.kth.se _archives_ volume dumps into TSM by dumping to a file
and then archiving the file.

MNAndrews>I'm not particularly familiar with ADSM, but we use HPSS(another HSM
MNAndrews>from IBM) at our site. 

Your place must have a lot of bucks, at least before the purchase of HPSS.

>(BTW: for linux only it would be very nicer if "afs" manages only the
>volumes, and passes all queries (kernel acl, quotas on server, ..) to
>the right server. the content of volumes stored like a local tree.)

A big part of AFS is the cache handling (and consistency) on the
client, and modifications to the files bypassing the cache manager
would invalidate the consistency. There are features in AFS (like
volume cloning) that would not fit very well to the structure you