Audit my RAID configs, please?

michaelsanford · 01-28-2008, 06:37 PM

My configuration:
I have done this before, but only once. Given this, I wanted to be sure of my setup before I started formatting system hard disks.

I have four SATA hard drives: 2(120) + 2(500) GB.

I want to use my two large SATA drives in a RAID 0 array for 1 TB of straight, concatenated storage space for non-mission-critical data, on a mount point /data

I want to use my two 120 GB drives in a RAID 1 array, onto which will be installed my operating system (Slackware 12), MySQL databases, and web site data (most of it, anyway). Very low traffic but need good availability, on /.

Unless there's a very good reason to partition my RAID 1 array disks into /srv, /local, /etc, etc, I don't want to bother (but please mention it if there is a good reason).

My protocol:
1. Separate my SATA drives onto the two channels available on my system board (so that both RAID arrays are represented on each channel, saving overhead on the RAID 1 array writes).

2. Reboot to the Slackware install CD.

3. Format my drives accordingly (meaning, one large partition on each disk of type Linux Raid Autodetect, with the addition of a 4 GB Linux Swap partition on both 120 GB drives). The bootable flag will also be set on /dev/sda1 (but not /dev/sdc1, also, right?)

4. Create /etc/raidtab with (bold means I don't know what this is for, or if it should be changed from my guide):

Code:

raiddev                 /dev/md0
# System RAID array of 120 GB disks
raid-level              1
nr-raid-disks           2
nr-spare-disks          0
persistent-superblock   1
device                  /dev/sda1
raid-disk               0
device                  /dev/sdc1
failed-disk             1
chunk-size		32

#/data array of 500 GB disks
raiddev                 /dev/md1
raid-level              1
nr-raid-disks           2
nr-spare-disks          0
persistent-superblock   1
device                  /dev/sdb1
raid-disk               0
device                  /dev/sdd1
failed-disk             1
chunk-size		32

(N.B. the drives' /dev assignments are staggered because the drives are not on the same channel)

5. Then make the RAID arrays

Code:

mkraid /dev/md0
mkraid /dev/md1

6. Make ReiserFS filesystems

Quote:

mkreiserfs /dev/md0
mkreiserfs /dev/md1

7. Install Slackware onto device /dev/md0

8. Choose to mount /dev/md1 as /data at boot time (I think it's an option in Setup).

9. Configure Lilo to mount /dev/md0 and install (in the MBR, superblock, where?)

Anything particularly wrong with this setup? I'm most worried about choosing where to install the lilo configuration (#9).

Thanks!

Electro · 01-28-2008, 06:57 PM

Yes, there is something wrong. You are using ReiserFS. I strongly recommend do not use ReiserFS. ReiserFS has reliability issues. It is best to use EXT3, XFS, or JFS.

Also you said you want 1 TB from two 500 GB hard drives in RAID-0. Well you are using RAID-1 for those hard drives which is half that you want. Putting two hard drives in RAID-0 with that capacity is dangerous. You are better off using Western Digital 1 TB or Hitachi 1 TB in RAID-1. If you are going to use 500 GB to reach to a terabyte, use four in a RAID-10. This will be safer if a hard drive goes. Though two drives could go, but they have to be adjacent to each other.

I suggest use GRUB instead of LILO.

I suggest spending a little more on hard drive models with the least amount of latency or accessing time for a server. Western Digital 'Raptor' is good and very fast.

michaelsanford · 01-28-2008, 07:10 PM

Quote:

Originally Posted by Electro

Also you said you want 1 TB from two 500 GB hard drives in RAID-0. Well you are using RAID-1 for those hard drives which is half that you want. Putting two hard drives in RAID-0 with that capacity is dangerous. You are better off using Western Digital 1 TB or Hitachi 1 TB in RAID-1. If you are going to use 500 GB to reach to a terabyte, use four in a RAID-10. This will be safer if a hard drive goes. Though two drives could go, but they have to be adjacent to each other.

I'm not using a redundant raid-level on the /data array specifically because I don't care about fault tolerance on that array. It contains data that is so non-mission-critical that the whole thing could go and it wouldn't matter much. What matters to me is the system and my web sites, which is why that 116 GB array (of two 120 - 4 GB for swap space) is RAID 0.

Also, my server is a modest tower, not a rack, and I have maxed out my hardware as it is (it only has 4 SATA connectors on the mainboard, for example). The function of my /data RAID 0 array is to serve as a NAS, for a very small LAN, and little more. Spending cash for a 1 TB hard disk is not in my cards...it just doesn't suit my purpose.

But thanks for the comments!

Electro · 01-29-2008, 03:34 AM

FYI, Linux does not use the BIOS of the storage controller to calculate the capacity of the hard drive. Linux calculates the geometry of the hard drive by it self. If the controller can not handle a terabyte hard drive, Linux is able to do it by by-passing the BIOS of the storage controller.

I think you meant RAID-1 for the to 120 GB array instead of RAID-0.

If it were me, I would be using two RAID-1 arrays and place web on one and mysql on the other. Also by spending on hard drives with lower accessing times, the data can be retrieve much faster than the drives that you selected. Make sure you use ECC memory.

On-board controllers are not reliable and Linux does not support hot-swapping for all on-board controllers.

michaelsanford · 01-29-2008, 11:58 AM

Yes I did mean RAID 1 for the 120 array... I always get that wrong; I need to come up with a mnemonic for that.

And thanks for the tip!