[GLLUG] Suggestions on hardware migration?
STeve Andre'
andres at msu.edu
Tue Jul 12 16:18:32 EDT 2005
I guess I should have said for a filesystem with lots of stuff on it, but
I so often am at 95%+ capacity it's normal for me.
I'm not a Linux user, but in my experience with SunOS and varibous
BSD varients, I've not had disk geometry problems with identical
disks. But they do want to be identical.
I'll have to look at the discussion of cp and tar for copying; I've stayed
with the tar method when cp just pain died when copying hundreds
of gigs of data in thousands of directories. But obviously there are
tens of ways to do anything.
--STeve Andre'
On Tuesday 12 July 2005 15:55, Matt Graham wrote:
> On Tuesday 12 July 2005 15:19, after a long battle with technology,
>
> STeve Andre' wrote:
> > On Tuesday 12 July 2005 13:39, Matt Graham wrote:
> > > On Tuesday 12 July 2005 12:51, after a long battle with technology,
> > >
> > > Lachniet, Mark wrote:
> > > > So what I was thinking is to just do a dd copy (e.g. 'dd
> > > > if=/dev/hda of=/dev/hdc bs=512') to just move the whole thing
> > > > over intact
> > >
> > > Nope. For one thing, you never use bs=512 for moving large amounts
> > > of data. For another thing, dd is the worst backup method in
> > > existence. You should hook both disks up to the same machine,
> > > cfdisk the new disk, mke2fs (or mkreiserfs) the partitions on the
> > > new disk, then cp -a each filesystem.
> >
> > Um, dd is the fastest possible way to clone disks,
>
> dd copies *everything*, including the parts of the disk that aren't
> actually being used. If your filesystems have significant chunks of
> empty space on them, dd does more I/O than you need. I/O is slow.
>
> > But dd can take raw devices and scoop up large quantities of disk,
> > which really is the fastest way to *clone* a disk.
>
> There are many gotchas with dd. Since "disk geometry" is still
> important on the x86, and 2 otherwise identical disks can report
> differing disk geometries depending on jumper positions and the phase
> of the moon, it's not as reliable as filesystem-based methods.
>
> > Doing a 'cp -a' won't keep permissions, unless -a implies something
> > different on Linux than BSD systems.
>
> Yes. Linux uses GNU cp, where -a is short for --archive and is
> equivalent to -dpPR in non-GNU cp.
>
> > When I want to move all the files from one partition to another I use
> > tar:
>
> This approach is really a hangover from the days when cp didn't always
> handle device files, symlinks, sockets, FIFOs, and so forth properly.
> There was recently a col.misc thread on the tar vs. cp thing;
> http://groups.google.com/advanced_group_search , Message-ID
> 1120910252.012174.291690 at g44g2000cwa.googlegroups.com for what several
> people had to say on the subject. GNU cp handles special files
> properly; BSD cp should do that by now but you never know what those
> crazy BSDers have been smoking :-) .
>
> > dd on the other hand is the best possible tool to clone two exactly
> > similar disks. For that purpose nothing can be faster.
>
> Depends on things. If you have an 80G disk with 78G of files on it,
> cloning the disk with dd may actually be faster than using
> filesystem-based methods. If you have an 80G disk with 40G of files on
> it, cloning via filesystem-based methods will be almost twice as fast
> as using dd.
More information about the linux-user
mailing list