[378] in Info-AFS_Redistribution
[NIK%ZURLVM1.BITNET@pucc.princeton.edu: AFS backup system]
daemon@ATHENA.MIT.EDU (sandy@alw.nih.gov)
Thu Nov 7 15:37:24 1991
Date: Thu, 7 Nov 91 13:19:30 EST
From: sandy@alw.nih.gov
To: NIK%ZURLVM1.BITNET@pucc.princeton.edu
Cc: Info-AFS@transarc.com
Excerpts from mail: 07 November 1991, 15:32:26 <Michael Niksch> [Subject: AFS backup system]
> To summarize my questions:
>
> 1) Is anybody out there ACTUALLY USING the AFS backup system at all ?
Cell 'alw.nih.gov` at the National Institutes of Health has been running a
production cell for several years using a backup system based on Transarc's
'backup` and 'butc` programs.
>
> 2) Has anybody managed to automate it enough to meet our requirements ?
>
We have found the Transarc backup system to be inadequately designed to handle
backups in a productions environment. The principle problems are:
1) inadequate support for scheduling backups
2) limited support for concurrent taping
3) unreliable protection of the database in certain circumstances
4) not robust in the face of connectivity and hardware failures
5) complex user interface
In an attempt to control these problems we have developed a backup system
using a shell script envelope for Transarc's software. The first version
has been in use through AFS3.0-AFS3.1. It provides
* an integrated backup cloning procedure
* facilities to schedule backups in full/incremental cycles
* some facilities for identifying and logging backup failures
* a backup administration report generator
* simplfied operations interface
* some protection of the database (ours, not theirs)
This system serviced about 30-50 users and 8-10 Gigabytes for several years.
Although the unreliable behavior of the 'butc` program prevents our
system from being fully robust, in practice it served us well and never
failed to produce restorable data when called upon, including two
full server disk recoveries. This system does require operator attention
to mount each tape.
In order to scale things up, we have rewritten our shell envelope as a
large Korn shell system which has these additional functions:
* schedules backups for more than one tape drive
* multiplexes on line requests for data recovery concurrent with backup
* simplifies system configuration in one administration file
* modular structure for easier maintenance
We have just completed development of this system and it has undergone
some testing. We plan to move over to it soon.
This software is in the public domain and can be made (but is not yet)
available to other cells through Transarc's "Grand Central Org" area.
However, documentation and other forms of support would be very limited.
> 3) If so, what amounts of data do these systems handle ?
We currently support about 600 volumes holding about 15 Gigabytes. By
distributing full backups over the work week we keep the daily load
at 10 Gigabytes.
>
> 4) How many tape drives can you use per machine / per cell ?
>
We have used three tape drives concurrently.
> 5) Has anybody tried to use a tape robot ?
No.
>
> 6) Does the AFS backup system compress data when dumping volumes ?
Not that I know of.
>
> 7) Would you recommend to just use vos dump, then handle the resulting
> disk files by some other backup and tape control system ?
I had written a system several years ago that used vosdump to write
data into a buffer partition which was then written by UNIX dump to tape.
This scheme worked well but was slower than the model used by Transarc
for writing directly to tape.
>
> 8) If so, can you do incremental backups, as incremental vos dump is
> reported to be broken, and incremental vos restore does not exist ?
We have not had any problem with this. In fact we found that when the
expected full backup is missing the incremental backup does the right
thing and dumps all data.
>
> 9) Would you recommend to just run tar on the AFS filesystem with a
> system:administrators token ?
>
I wouldn't use it.
> 10) Is anybody out there running AFS as a service (not just a playground)
> for 200+ users without regularly fixing and recompiling AFS source
> code himself ?
>
We are running a production system. We currently support over 200 AFS users
served by 5 files servers and over 20 Gigabytes of disk capacity.
> 11) If not, how far would you get with the export version (no DES license)
> and without an AIX source license ?
We have a source license for AFS, but have not entered into any software
modification for the backup system. Because of limited staffing,
we have had to rely on Transarc to fix bugs.
Sandy Orlow (Systex, Inc.)
Building 12A, Room 2013 phone: (301) 402-1533
National Institutes of Health uucp: uunet!sandy%alw.nih.gov
Bethesda, MD 20892 Internet: sandy@alw.nih.gov
Nurul Hamsath (Systex, Inc.) phone: (301) 496-0331
Building 12A, Room B57 uucp: uunet!nih-csl!nurul
National Institutes of Health Internet: nurul@alw.nih.gov
Bethesda, MD 20892
