Boston Linux & Unix (BLU) Home | Calendar | Mail Lists | List Archives | Desktop SIG | Hardware Hacking SIG
Wiki | Flickr | PicasaWeb | Video | Maps & Directions | Installfests | Keysignings
Linux Cafe | Meeting Notes | Blog | Linux Links | Bling | About BLU

BLU Discuss list archive


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RAID--quick, cheap, good/fast, pick all 3



As promised in yesterday's article, I am following up with a writeup about
RAID.  In the past the "I" in RAID was just a figure of speech, but now it
really *is* inexpensive.

Scenario:  My old Pentium-II box came with a 10-gig IDE drive, and I had a
blank 80-gig unit lying around unused since getting a warranty replacement in
the mail sometime last year.  I did my Suse 8.2 installation on the 80-gig
unit, taking care to put system and personal files in a 9-gig partition.

Inspiration:  the Software-RAID HOWTO at
http://www.tldp.org/HOWTO/Software-RAID-HOWTO.html .

Comments:
In the past I've always assumed that if you wanted decent performance, you had
to buy custom RAID hardware at fairly big prices.  In fact my home file server
is a three-kilobuck system I bought 3-1/2 years ago from a company that
specializes in turnkey 3Ware-based Linux boxes.  However, the HOWTO points out
what now seems to me pretty obvious:  any old motherboard has two separate IDE
buses, and a PCI bus outperforms hard drive throughput by more than a factor
of two.  So assuming you have a modern CPU, you can configure RAID without
buying any hardware at all, and get performance within a hair's breadth of a
typical hardware RAID controller.  In fact you can probably get quite good
performance even with a pretty old CPU.

And I also assumed that you needed multiple identical disk drives.  Not true
with software RAID.

Why RAID?  Well, it's a real time-consuming pain in the butt to reinstall a
system from backup.  And most of us are kinda lazy about backups.  Disc drives
are 100 times more reliable than they were 10 years ago--but do you want to
trust everything to a drive, especially if you can have a continuous automatic
backup with a mere 60 minutes or so of advance prep?

I knew nothing about the MD device driver at 5pm today.  At 10:30pm, I had my
system up and running in a RAID-1 configuration (spent three of those hours
bowling).  If I can do this in 60 minutes, so can you.

Steps to take:

(I used "method 2" in the RAID-1 section of the HOWTO.)

- Make sure you have a kernel built to support raid1.
- Install a drive on each IDE bus.  In my case hda is the 80GB drive, hdc is
the 10GB drive.
- Back up the smaller drive if you need anything from it anymore.
- Create a partition on the smaller drive equal in size or slightly smaller
than your current root partition.  (My root dir is 9GB on each drive.) Set its
type to "fd", for RAID autodetect.
- Using yast2 or an editor, create /etc/raidtab with the following contents:

--
raiddev /dev/md0
   raid-level       1
   nr-raid-disks    2
   nr-spare-disks   0
   persistent-superblock 1
   chunk-size        16
   device   /dev/hdc2
   raid-disk 0
   device   /dev/hda2
   raid-disk 1
   failed-disk 1
--
  In the above example, hda2 is the current root partition, and hdc2 is the
new partition.  (My new distro is on the 80GB drive /dev/hda, my old system
was on /dev/hdc--I'd made a copy of the old system first, and then wiped the
drive.)

- Invoke 'mkraid /dev/md0' to write the superblock to the new partion. (Takes
only a second.)
- Create a filesystem on /dev/md0 using mkfs.  (A reiserfs filesystem takes
only a couple seconds to generate.)  Mount it to /mnt.
- Shut down apps or go standalone, and copy your root filesystem to /dev/md0. 
I used the command 'cd / ; tar cf - . | (cd /mnt ; tar xpf - )'.  This is the
only lengthy part of the process, figure on a coffee break depending on how
big your root fs is.
- Update your lilo or grub configuration, and fstab in /mnt/etc, to point root
at /dev/md0 instead of its old setting.
- Reboot and you're running in degraded RAID1 on /dev/md0.  Once you're
satisfied with the system, run fdisk to change the partion type to 'fd' on
your former root, edit out the "failed-disk" line in /etc/raidtab, reboot
again and you'll see in /var/log/messages that both discs are now members of
the array but that one is unavailable.
- Invoke 'raidhotadd /dev/md0 /dev/hda2' to sync the two drives.  You'll see a
bunch of messages in /var/log/messages, and in a matter of minutes the two
drives will be synced.  But you won't need to go grab a cup of coffee this
time--you can go on your merry way as soon as you invoke raidhotadd.

Repeat the above with any other partitions you want beyond the root.

Bottom line:
My 5-year-old 10-gig drive has a new lease on life as a hot backup.  I no
longer have to entrust files to a new disk that may have sudden infant death,
nor do I have to entrust them to an old disk that might be near the end of its
life.  Peace of mind!

As for performance, do a couple of 'time cp foo bar' commands with say 100-meg
files before and after your upgrade.  I challenge you to measure the
difference, it's basically the same on my system.

Cost:
Zero.  The price of a disk upgrade that I had already done.  Labor:  an hour,
about the time you'd wait for a new pair of glasses ;-)

-rich




BLU is a member of BostonUserGroups
BLU is a member of BostonUserGroups
We also thank MIT for the use of their facilities.

Valid HTML 4.01! Valid CSS!



Boston Linux & Unix / webmaster@blu.org