[Dirvish] Copying banks

Per Marker Mortensen per at markermortensen.dk
Sun Jan 31 11:12:49 UTC 2010

At Wed, 20 Jan 2010 15:31:19 +0100,
Bernd Haug wrote:


> What I tried:
> $,1s"(B see above
> $,1s"(B the BSD dump port - it needs too much temp space (i.e., multi-GiB)
> for filesystems with a large directory structure and is also quite
> slow

I see others have suggested tar or dump, which preferably should be
run pipeing directly from input to output, perhaps using netcat or ssh.

I have myself been using cpio when moving LARGE amounts of dirvish
vaults with LOTS of hardlinks (fully linked using faster-dupemerge).

Found this old script on one of my servers:

# move old splint backups to new disks

cd /marker1/old/splint1-vservers

for DIR in `ls -1d 200?????-2250 | sort -nr`;
  echo $DIR
  find $DIR -depth -print | cpio -oaV -H newc | ssh root at splint.anc.dk
"cd /marker4/splint/splint1-vservers && cpio -imVd -H newc"
  mv $DIR $DIR.copied
  sleep 10
  ssh root at splint.anc.dk "cd /marker4/splint/splint1-vservers &&
faster-dupemerge --verbose --find -size +2000k --"

cd -

Seems I run cpio through ssh, and run a faster-dupemerge afterwards to
make sure everything is hardlinked correctly.

I seem to remember this took really long time, buts it does not require
large amounts of memory or disk to run.

Anyone else experimented with using cpio to move dirvish vaults?


Per Marker Mortensen <per at markermortensen.dk>

More information about the Dirvish mailing list