ADSM-L

Re: ADSM for AFS and DFS

1999-03-19 05:19:13
Subject: Re: ADSM for AFS and DFS
From: Peter Dümpert <c0034003 AT RZIBM214.RZ.TU-BS DOT DE>
Date: Fri, 19 Mar 1999 11:19:13 +0100
On Thu, 18 Mar 1999, S W Luan wrote:

> (1)  When using dsmcafs, dsmcdfs, is a large number (sometimes tens of 
> thousands) of VIRTUALMOUNTPOINT definitions for all the filesets/volumes
> required?  How might one be able to speed up the file backup process?
>
> There is a way to do per-volume (or per-fileset in DFS terminology)
> VIRTUALMOUNTPOINT without having to define/maintain them explicitly.  A
> script can be written to list all the volumes in the cell and mount them
to
> a selected directory (e.g., /afs/almaden.ibm.com/.adsm).  You use the
> environment-variable support in ADSM in the dsm.sys file to define a
> generic VIRTUALMOUNTPOINT.  For example,
>
> VIRTUALMOUNTPOINT ${AFSVOLUME}
.....

Point (1) of Luan's remarks are exactly what we do.

Based upon an idea, Heidelberg University Computing Centre
        http://web.urz.uni-heidelberg.de/ADSM/
uses since early ADSM-days (called WDSF in these days, running an ADSM-server 
on a VM-machine), our scenario is as follows:

Environment:
AFS-servers running on HP-machines under HP-UX 10.20
AFS-client-machines under HP-UX 10.20
AFS-client on a RS6K C10, 64 MB, AIX 4.2.1 ( 4.3.2 soon )
  dsmcafs (V3.1.0.3) backing the whole of AFS user-$HOMEs on FILE-basis
  dsmbuta for an additional volume backup
ADSM-server (V3.1.1.5) under AIX 4.2.1 ( 4.3.2 soon ) on a F40, 256 MB
  Dual-Processor         

Backup:
User-$HOMEs are in AFS
our directory-structure is
        /afs/tu-bs.de/usr/usrxx/individual_user
  the "usrxx" runs from "usr0" to "usr99", i.e. xx = UID modulo 100
  ( The UID being the numerical uid of a user )
We've about 15000+ VirtualMointPoints (VMPs)
The VMPs are addressed in the dsm.sys AND dsm.opt files via ${USER_DIR}
  (dsm.opt : DOMAIN ${USER_DIR}, dsm.sys : VIRTULMountpoint ${USER_DIR} )
This USER_DIR env-variable is set in a script per "dsmcafs"-call, i.e.
  we run each night 15000 "dsmcafs" calls, i.e. one call per user.
in this big loop we check wether the resp. afsvolume was accessed within
  the last 3 days, i.e the # of actually active afsvolumes is about 10% 
  of the 15000 afsvolumes and ONLY these ca. 1500 are really backup'd.
currently we run 4 streams in parallel based upon our tree-structure
  (0 <= xx <= 24) (25 <= xx <= 49) (50 <= xx <= 74) (95 <= xx <= 99)
  on a relatively small client-machine ( IBM C10, 64MB ) lasting ca.
  4 - 7 hours per stream ( depending on the weekday), so NO time-problem
  with a file-backup. 
  How many streams You actually run very much depends on the capabilities
  of the ONE ( YOU need 1 license !! ) client-server machine, i.e. a
  stronger machine than e.g. our C10 could easily run more than 4 streams
  in parallel.

NOW the restores:  
  
again taking the idea of Heidelberg University + the ENV-variable-
  facility in dsm.opt + dsm.sys the individual user runs a "dsmc"-script
  which does an
        export USER_DIR=$HOME
  The ${USER_DIR} being used in the dsm.opt AND dsm.sys files
  (see backup above )
additionally the "dsmcafs" is called e.g.
        dsmcafs restore -server=client_machine_which_does_the_backup ...
the DOMAIN is recognised properly by "dsmc/dsmcafs" BUT NOT by the GUI
  "dsm/dsmafs".
  We therefore recommend the "line"-interface for real work.
  The GUI presents the above mentioned 15000 VMPs, BUT if You train the
  user to use the "search"-Button, the GUI presents a popup-window, where
  the starting-point is $USER_DIR.
this works BOTH for HP-UX-clients (/usr/adsm/dsmc) AND AIX-AFS-clients 
  (/usr/lpp/adsm/bin/dsmc /usr/lpp/adsm/bin/dsmcafs )
So, if we don't bother with AFS-ACL's ANY user on HP-client machines
  can do restores. If a user needs the ACL's we currently tell him to use
  an AIX-box ( using the dsmcafs )

Remarks on ADSM+ARCHIVE:

we don't use ADSM's ARCHIVE yet and probably won't use it in the future,
  because we feel ADSM is still not yet settled down for LONG TERM
  ( > 5 years ) archiving. A further point is the lack of accessing
  an archived file in an emergency case ( i.e. no ADSM-DB available ).
we therefore intend to use a product from LEEDS University, UK

        http://www.leeds.ac.uk/ucs/systems/archive.html
        http://www.leeds.ac.uk/ucs/documentation/how/index.html
                        ( HOW 14 How to archive files )
        http://www.leeds.ac.uk/ucs/systems/archive/
        http://www.leeds.ac.uk/ucs/systems/archive/sysprog.html

  which we think addresses the needs for university users by far better
  than ADSM does.

-- 
MfG / Ciao         - - - - - - - - - - - - - - - - - - - - - - - - - - - -
MfG / Ciao         - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Peter Dümpert                                   Email: p.duempert AT tu-bs DOT 
de
Rechenzentrum der Technischen Universität       Fax  : ++49/531/391-5549
D 38092 Braunschweig                            Tel  : ++49/531/391-5535
 
<Prev in Thread] Current Thread [Next in Thread>