[Dirvish] Copying dirvish vaults (heavily hard linked)

[Shawn (Red Mop)]

On Thursday 06 March 2008 11:28:31 am Keith Lofstrom wrote:
> On Thu, Mar 06, 2008 at 09:33:52AM -0700, Shawn (Red Mop) wrote:
> > Tar is actually very fast.  I used it to split a bank about a month ago. 
> > It was a little unnerving, so I ran rsync -aH on top of (after) tar, and
> > rsync did almost nothing.  The du -hcs * in each vault already matched.
>
> "Almost nothing", unfortunately, can be a big difference.  When I was
> testing, tar did not produce bootable images, because some of the special
> files did not copy.   Perhaps it has been improved.  rsync does produce
> bootable images (if it did not, then dirvish wouldn't be very useful).
>
> My point is that these file copy tools can take a surprisingly long
> time.  My 500GB backup drive is now about 70% full, with about 100
> images on it (I do not expire them), meaning that to some tools, on
> the file level, the drive appears to contain around 60 terabytes.
> A drive copy at the file level will attempt to traverse a large
> fraction of that 60TB;  if the interface is running at 20MB/sec
> that will take a large fraction of 800 hours.  Yikes!
>
> > LVM + resize2fs is one of my very best friends.  I grow and shring
> > partions all the time.  It's a real kick to move a partition from one
> > disk to another while the partition is in use without rebooting, or even
> > taking the partition offline.
>
> Yes, fine tools.  I wish resize2fs had an extra option for increasing
> or decreasing inode *density*.  It does add or subtract inodes as you
> grow or shrink partitions, maintaining the ratio of inodes to data,
> but sometimes the problem is a lack or surplus of inodes.  "Plan
> ahead" is not always possible or easy with heavily hard-linked file
> systems with unpredictable future usage.
>
> Is there a tool that permits the manipulation of inode density on
> a used file system?
>
> Keith

I was worried about booting too.  Thus, I did run rsync after.  It did run 
faster running tar then rsync over just rsync.  Perhaps I should have added 
that.