Re: What's the reasonable time to mirror a hard disk?
hi ya yuwen
how full is /dev/hda ???
- if some of your partitions is 75% full... i would NOT use dd...
( faster to do it manually
assuming that /var and /usr not too full but / is 80% full etc
check the dma status of your disk
root# hdparm /dev/hda
root# hdparm /dev/hdb
to see the write/read xfer speeds
root# hdparm -tT /dev/hda
turn dma mode on if its not set
root# hdparm -d1 /dev/hda
fdisk your /dev/hdb the same way or different way than /dev/hda
-- remember that dd keeps the same original size partition
now copy your directories in each parititons...
dd if=/dev/hda1 /dev/hdb1 bs=32k ( / )
# or #
( tar cf - /bin /sbin /lib /boot ) | ( cd /mnt/hdb1 ; tar xvfp - )
mount /dev/hdbxxx /mnt/xxx
( tar cf - /var ) | ( cd /mnt/hdb2 ; tar xvfp - )
( tar cf - /usr ) | ( cd /mnt/hdb3 ; tar xvfp - )
-- add/delete partions as needed per your disk
-- add missing dirs ... /tmp /.automount /.autofs ...
-- you should be done within 1hr.... on celeron-500....
15-30 min on P3-800
-- chroot and fix lilo if you wanna boot off of the new /dev/hdb
c ya
alvin
i'd do the following
On Mon, 22 Oct 2001, Noah Meyerhans wrote:
> On Mon, Oct 22, 2001 at 03:44:28AM +0000, Yuwen Dai wrote:
> > I have two identical hard disk linked with one cable. The capacity of each
> > disk is 40G. I want to have the second disk be the mirror of the first disk
> > by using this command:
> >
> > dd if=/dev/hda of=/dev/hdb bs=32k
> >
> > Nearly 2 hours passed, dd still hadn't finished. I had to press 'Ctrl-C' to
> > stop it. Is this the right way to mirror a disk? Or is there some better
> > way to do this? Thanks in advance.
>
> First of all, you'd probably be better off using dump and restore rather
> than dd. dd will copy a block even if it's completely unused by the
> filesystem. So if the drive you're backing up is only 25% full, then dd
> will take 4 times longer than dump, simply because it's copying a bunch
> of unnecessary blocks.
>
> Also, you're probably better off moving these drives to separate
> controllers. I don't remember all the details of it, but IDE has
> traditionally had a limitation that allowed a controller to only queue
> commands for a single drive at a time. Such a limitation would have a
> major impact on performance in a case like this. Unfortunately I can't
> provide you with a reference for this, and I don't know if it's still
> the case with modern IDE controllers.
>
> noah
>
> --
> _______________________________________________________
> | Web: http://web.morgul.net/~frodo/
> | PGP Public Key: http://web.morgul.net/~frodo/mail.html
>
Reply to: