[OpenAFS] Re: Current OpenAFS Backup Recommendations

[Adam Megacz]

Russ Allbery <rra@stanford.edu> writes:
>> You can get remarkably good compression by using xdelta3 on dump
>> files.

> Don't you have to do full dumps in order to calculate the xdelta3
> differences?

I pipe the output of "vos dump" directly to xdelta3, so the full dump
never hits the disk.  If you perform the dump and diff on the
fileserver itself, then the network traffic is also proportional to
the size of the diff (rather than the size of the whole dump).  You
can then send the diff over the network to secondary storage.
Secondary storage needs to keep a copy of the full dump and reapply
the diff, so the amount of effort is proportional to the size of the
dump, but the network traffic is proportional to the diff.

It takes a few tries to set it up, but works really smoothly once
you've got it going.  IMHO it's the best incremental backup solution
for AFS that doesn't involve the risk of your proprietary software
suddenly becoming unsupported.

The ideal situation would be to have the volserver emit an RFC3284
file directly using its knowledge of which file blocks have changed
since a given date.  This would give you block-granularity incremental
dumps instead of the file-granularity dumps without inventing a new
dump format; just use an RFC3284-of-dumpfiles.

  - a