Coda File System

Re: Backup questions

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Fri, 6 Apr 2001 15:28:34 -0400
On Thu, Apr 05, 2001 at 05:25:25PM +0200, Steffen Neumann wrote:
> Hi,
> 
> By now we have a prototype backup solution in place,
> using the coda-provided scripts and tools and a 
> standalone SCSI Tape.
> 
> Ideally we'd like to shove the workload to our
> computing support group, who do backups using 
> a Veritas NetBackup solution.
> 
> However, they're a little reluctant 
> to move the volume files as full dumps 
> (even if they're incrementals themselves), 

Why would it matter whether "dump" or "volutil dump" produces a blob of
data (either full or incremental).

> Apart from continuing the standalone system:
> 
> would it be feasable to use an afs-veritas client ?
> Of course that depends on their implementation,
> which I don't know about. 

I wouldn't expect that to be feasable, different communication
protocols, very different metadata (AFS doesn't have versionvectors).

> The Amanda backup, as far as I got it, also lack file indexing. Things
> are a two-level theme, still. Is that correct ?

Yeah, there is as yet no utility that can pull a single file out of a
volume dump, so there is not much use for file indexing. However, most
accidental file losses are covered by mounting the backup volume under
"OldFiles".

F.i. when /coda/usr/jaharkes is the volume "vmm:u.jaharkes",

    cfs mkm /coda/usr/jaharkes/OldFiles vmm:u.jaharkes.0.backup

The backup volume is 're-cloned' everytime backups are done. So it would
normally contains yesterdays files. This saves a lot of going back to
tape for a single file situations in case of user error.

The case of losing a complete server is either solved by resolving with
other replicas (again no need for tapes), or restore full dumps anyways.

Jan
Received on 2001-04-06 15:30:00