From ArchWiki
Revision as of 00:17, 8 October 2013 by Graysky (Talk | contribs) (Add to Kernel Image: removed noise - udev will detect modules)

Jump to: navigation, search

Template:Article summary start Template:Article summary text Template:Article summary heading Template:Article summary wiki Template:Article summary wiki Template:Article summary wiki Template:Article summary end


Template:Wikipedia Redundant Array of Independent Disks (RAID) is a storage technology that combines multiple disk drive components (typically disk drives or partitions thereof) into a logical unit. Depending the RAID implementation, this logical unit can be a file system or an additional transparent layer that can hold several partitions. Data is distributed across the drives in one of several ways called "RAID levels", depending on the level of redundancy and performance required. The RAID level chosen can thus prevent data loss in the event of a hard disk failure, increase performance or be a combination of both.

Despite redundancy implied by most RAID levels, RAID does not guarantee that data is safe. A RAID will not protect data if there is a fire, the computer is stolen or multiple hard drives fail at once. Furthermore, installing a system with RAID is a complex process that may destroy data.
Warning: Therefore, be sure to backup all data before proceeding.

There are many different levels of RAID, please find hereafter the most commonly used ones.

Standard RAID levels

Uses striping to combine disks. Even though it does not provide redundancy, it is still considered RAID. It does, however, provide a big speed benefit. If the speed increase is worth the possibility of data loss (for swap partition for example), choose this RAID level. On a server, RAID 1 and RAID 5 arrays are more appropriate. The size of a RAID 0 array block device is the size of the smallest component partition times the number of component partitions.
The most straightforward RAID level: straight mirroring. As with other RAID levels, it only makes sense if the partitions are on different physical disk drives. If one of those drives fails, the block device provided by the RAID array will continue to function as normal. The example will be using RAID 1 for everything except swap and temporary data. Please note that with a software implementation, the RAID 1 level is the only option for the boot partition, because bootloaders reading the boot partition do not understand RAID, but a RAID 1 component partition can be read as a normal partition. The size of a RAID 1 array block device is the size of the smallest component partition.
Requires 3 or more physical drives, and provides the redundancy of RAID 1 combined with the speed and size benefits of RAID 0. RAID 5 uses striping, like RAID 0, but also stores parity blocks distributed across each member disk. In the event of a failed disk, these parity blocks are used to reconstruct the data on a replacement disk. RAID 5 can withstand the loss of one member disk.
Note: RAID 5 is a common choice due to its combination of speed and data redundancy. The caveat is that if one drive were to fail before and that drive was replaced another drive failed too, all data will be lost.

Nested RAID levels

RAID 1+0
Commonly referred to as RAID 10, is a nested RAID that combines two of the standard levels of RAID to gain performance and additional redundancy. It is the best alternative to RAID 5 when redundancy is crucial.

RAID level comparison

RAID level Data redundancy Physical drive utilization Read performance Write performance Min drives
0 No 100% nX




1 Yes 50% nX (theoretically)

1X (in practice)

1X 2
5 Yes 67% - 94% (n−1)X




6 Yes 50% - 88% (n−2)X (n−2)X 4
10 Yes 50% (n−2)X (n−2)X 4

* Where n is standing for the number of dedicated disks.


The RAID devices can be managed in different ways:

Software RAID
This is the easier implementation as it does not rely on obscure proprietary firmware and software to be used. The array is managed by the operating system either by:
  • by an abstraction layer (e.g. mdadm);
    Note: This is the method we will use later in this guide.
  • by a logical volume manager (e.g. LVM);
  • by a component of a file system (e.g. ZFS).
Hardware RAID
The array is directly managed by a dedicated hardware card installed in the PC to which the disks are directly connected. The RAID logic runs on an on-board processor independently of the host processor (CPU). Although this solution is independent of any operating system, the latter requires a driver in order to function properly with the hardware RAID controller. The RAID array can either be configured via an option rom interface or, depending on the manufacturer, with a dedicated application when the OS has been installed. The configuration is transparent for the Linux kernel: it doesn't see the disks separately.
This type of RAID is properly called BIOS or Onboard RAID, but is falsely advertised as hardware RAID. The array is managed by pseudo-RAID controllers where the RAID logic is implemented in an option rom or in the firmware itself with a EFI SataDriver (in case of UEFI), but are not full hardware RAID controllers with all RAID features implemented. Therefore, this type of RAID is sometimes called FakeRAID. dmraid from the official repositories, will be used to deal with these controllers. Some FakeRAID controller examples: Intel Rapid Storage, JMicron JMB36x RAID ROM, AMD RAID, ASMedia 106x,...

Which type of RAID do I have?

As the process to set up software RAID is completely user driven.

However, discerning between FakeRAID and true hardware RAID can be more difficult. As stated, manufacturers often incorrectly distinguish these two types of RAID and false advertising is always possible. The best solution in this instance is to run the lspci command and looking through the output to find the RAID controller. Then do a search to see what information can be located about the RAID controller. True hardware RAID controller are often rather expensive (~$400+), so if someone customized the system, and it is very likely that choosing a hardware RAID setup made a very noticeable change in the computer's price.


Install mdadm which is used for administering pure software RAID using plain block devices: the underlying hardware does not provides any RAID logic, just a supply of disks. mdadm will work with any collection of block devices. Even if unusual. For example, one can thus make a RAID array from a collection of thumb drives.

Prepare the Devices

Warning: These steps erase everything on a device, so type carefully!

To prevent possible issues each device in the RAID should be securely wiped. If the device is being reused or re-purposed from an existing array, erase any old RAID configuration information:

# mdadm --zero-superblock /dev/<drive>

Create the Partition Table

It is highly recommended to pre-partition the disks to be used in the array. Since most RAID users are selecting HDDs >2 TB, GTP partition tables are required and recommended. Disks are easily partitioned using gptfdisk.

  • After created, the partition type should be assigned hex code FD00.
  • Creating partitions that are of the same size on each of the devices is preferred.
  • A good tip is to leave approx 100 MB at the end of the device when partitioning. See below for rationale.
Note: It is also possible to create a RAID directly on the raw disks (without partitions), but not recommended because it can cause problems when swapping a failed disk.

When replacing a failed disk of a RAID, the new disk has to be exactly the same size as the failed disk or bigger — otherwise the array recreation process will not work. Even hard drives of the same manufacturer and model can have small size differences. By leaving a little space at the end of the disk unallocated one can compensate for the size differences between drives, which makes choosing a replacement drive model easier. Therefore, it is good practice to leave about 100 MB of unallocated space at the end of the disk.

Build the Array

Use mdadm to build the array. Several examples are given below.

Warning: Do not simply copy/paste the examples below; use your brain and substitute the correct options/drive letters!
Note: If this is a RAID1 array which is intended to boot from Syslinux a limitation in syslinux v4.07 requires the metadata value to be 1.0 rather than the default of 1.2.

The following example shows building a 2 device RAID1 array:

# mdadm --create --verbose --level=1 --metadata=1.2 --chunk=64 --raid-devices=2 /dev/md0 /dev/sdb1 /dev/sdc1
Note: In a RAID1 the chunk switch is actually not needed.

The following examples shows building a 3 device RAID5 array:

# mdadm --create --verbose --level=5 --metadata=1.2 --chunk=256 --raid-devices=3 /dev/md0 /dev/sdb1 /dev/sdc1 /dev/sdd1 --spare-devices=1 /dev/sde1
Tip: If the array is being created on new HDDs, one can avoid the initial resync by adding the --assume-clean flag.

The array is created under the virtual device /dev/mdX, assembled and ready to use (in degraded mode). One can directly start using it while mdadm resyncs the array in the background. It can take a long time to restore parity. Check the progress with:

$ cat /proc/mdstat

Assemble the Array

After built, default configuration file, mdadm.conf needs to be updated like so:

# mdadm --detail --scan >> /etc/mdadm.conf
Note: If updating the RAID configuration from within the Arch Installer by swapping to another TTY, ensure that the correct mdadm.conf file is written:
# mdadm --detail --scan >> /mnt/etc/mdadm.conf

Once the configuration file has been updated the array can be assembled using mdadm:

# mdadm --assemble --scan

Format the RAID Filesystem

The array can now be formatted like any other disk, just keep in mind that:

  • Due to the large volume size not all filesystems are suited (see: File system limits).
  • The filesystem should support growing and shrinking while online (see: File system features).
  • One should calculate the correct stride and stripe-width for optimal performance.

Calculating the Stride and Stripe-width

Stride = (chunk size/block size). Stripe-width = (# of physical data disks * stride).

Example 1. RAID1

Example formatting to ext4 with the correct stripe-width and stride:

  • Hypothetical RAID1 array is composed of 2 physical disks.
  • Chunk size is 64k.
  • Block size is 4k.

Stride = (chunk size/block size). In this example, the math is (256/4) so the stride = 16.

Stripe-width = (# of physical data disks * stride). In this example, the math is (2*16) so the stripe-width = 32.

# mkfs.ext4 -v -L myarray -m 0.5 -b 4096 -E stride=16,stripe-width=32 /dev/md0
Example 2. RAID5

Example formatting to ext4 with the correct stripe-width and stride:

  • Hypothetical RAID5 array is composed of 4 physical disks; 3 data discs and 1 parity disc.
  • Chunk size is 256k.
  • Block size is 4k.

Stride = (chunk size/block size). In this example, the math is 256/4) so the stride = 64.

Stripe-width = (# of physical data disks * stride). In this example, the math is (3*64) so the stripe-width = 192.

# mkfs.ext4 -v -L myarray -m 0.5 -b 4096 -E stride=64,stripe-width=192 /dev/md0

For more on stride and stripe-width, see: RAID Math.

Add to Kernel Image

Note: The following section is applicable only if the root filesystem resides on the array. Users may skip this section if the array holds a data partition(s).

Add mdadm_udev to the HOOKS section of the Mkinitcpio to add support for mdadm directly into the init image:

HOOKS="base udev autodetect block mdadm_udev filesystems usbinput fsck"

Be sure to regenerate the initramfs image (see Image creation and activation) after making the modification to the HOOKS array:

# mkinitcpio -p linux

Mounting from a Live CD

Users wanting to mount the RAID partition from a Live CD, use

# mdadm --assemble /dev/<disk1> /dev/<disk2> /dev/<disk3> /dev/<disk4>

RAID Maintenance


It is good practice to regularly run data scrubbing to check for and fix errors. Depending on the size/configuration of the array, a scrub may take multiple hours to complete.

To initiate a data scrub:

# echo check > /sys/block/md0/md/sync_action

The check operation scans the drives for bad sectors and automatically repairs them. If it finds good sectors that contain bad data (the data in a sector does not agree with what the data from another disk indicates that it should be, for example the parity block + the other data blocks would cause us to think that this data block is incorrect), then no action is taken, but the event is logged (see below). This "do nothing" allows admins to inspect the data in the sector and the data that would be produced by rebuilding the sectors from redundant information and pick the correct data to keep.

As with many tasks/items relating to mdadm, the status of the scrub can be queried by reading /proc/mdstat.


$ cat /proc/mdstat
Personalities : [raid6] [raid5] [raid4] [raid1] 
md0 : active raid1 sdb1[0] sdc1[1]
      3906778112 blocks super 1.2 [2/2] [UU]
      [>....................]  check =  4.0% (158288320/3906778112) finish=386.5min speed=161604K/sec
      bitmap: 0/30 pages [0KB], 65536KB chunk

To stop a currently running data scrub safely:

# echo idle > /sys/block/md0/md/sync_action
Note: If the system is rebooted after a partial scrub has been suspended, the scrub will start over.

When the scrub is complete, admins may check how many blocks (if any) have been flagged as bad:

# cat /sys/block/md0/md/mismatch_cnt

General Notes on Scrubbing

Note: Users may alternatively echo repair to /sys/block/md0/md/sync_action but this is ill-advised since if a mismatch in the data is encountered, it would be automatically updated to be consistent. The danger is that we really don't know whether it's the parity or the data block that's correct (or which data block in case of RAID1). It's luck-of-the-draw whether or not the operation gets the right data instead of the bad data.

It is a good idea to set up a cron job as root to schedule a periodic scrub. See raid-checkAUR which can assist with this.

RAID1 and RAID10 Notes on Scrubbing

Due to the fact that RAID1 and RAID10 writes in the kernel are unbuffered, an array can have non-0 mismatch counts even when the array is healthy. These non-0 counts will only exist in transient data areas where they don't pose a problem. However, since we can't tell the difference between a non-0 count that is just in transient data or a non-0 count that signifies a real problem. This fact is a source of false positives for RAID1 and RAID10 arrays. It is however recommended to still scrub to catch and correct any bad sectors there might be in the devices.

Adding Devices to an Array

Adding new devices with mdadm can be done on a running system with the devices mounted. Partition the new device using the same layout as one of those already in the arrays as discussed above.

Assemble the RAID arrays if they are not already assembled:

# mdadm --assemble /dev/md1 /dev/sda1 /dev/sdb1 /dev/sdc1
# mdadm --assemble /dev/md2 /dev/sda2 /dev/sdb2 /dev/sdc2
# mdadm --assemble /dev/md0 /dev/sda3 /dev/sdb3 /dev/sdc3

Add the new device as a spare device to all of the arrays:

# mdadm --add /dev/md0 /dev/sdxy

This should not take long for mdadm to do. Again, check the progress with:

# cat /proc/mdstat

Check that the device has been added with the command:

# mdadm --misc --detail /dev/md0

It should be listed as a spare device.

Tell mdadm to grow the arrays from 3 devices to 4 (or to however many devices are being added):

# mdadm --grow -n 4 /dev/md0

This will probably take several hours. Users must wait for it to finish before continuing. Check the progress in /proc/mdstat.

Removing Devices from an Array

One can remove a device from the array after marking it as faulty:

# mdadm --fail /dev/md0 /dev/sdxx

Now remove it from the array:

# mdadm -r /dev/md0 /dev/sdxx

Remove device permanently (for example, to use it individually from now on): Issue the two commands described above then:

# mdadm --zero-superblock /dev/sdxx
Warning: Reusing the removed disk without zeroing the superblock will case LOSS of all data on the next boot. (After mdadm will try to use it as the part of the raid array). DO NOT issue this command on linear or RAID0 arrays or data LOSS will occur!

Stop using an array:

  1. Umount target array
  2. Stop the array with: mdadm --stop /dev/md0
  3. Repeat the three command described in the beginning of this section on each device.
  4. Remove the corresponding line from /etc/mdadm.conf


A simple one-liner that prints out the status of the RAID devices:

awk '/^md/ {printf "%s: ", $1}; /blocks/ {print $NF}' </proc/mdstat
md1: [UU]
md0: [UU]

Watch mdstat

watch -t 'cat /proc/mdstat'

Or preferably using tmux

tmux split-window -l 12 "watch -t 'cat /proc/mdstat'"

Track IO with iotop

The iotop package displays the input/output stats for processes. Use this command to view the IO for raid threads.

iotop -a -p $(sed 's, , -p ,g' <<<`pgrep "_raid|_resync|jbd2"`)

Track IO with iostat

The iostat package displays the input/output statistics for devices and partitions.

 iostat -dmy 1 /dev/md0
 iostat -dmy 1 # all

Mailing on events

A smtp mail server (sendmail) or at least an email forwarder (ssmtp/msmtp) is required to accomplish this. Perhaps the most simplistic solution is to use dmaAUR which is very tiny (installs to 0.08 MiB) and requires no setup.

Edit /etc/mdadm.conf defining the email address to which notifications will be received.

Note: If using dma as mentioned above, users may simply mail directly to the username on the localhost rather than to an external email address.

To test the configuration:

# mdadm --monitor --scan --test

When satisfied with the results, enable the mdadm daemon:

# systemctl enable mdadm.service


If you are getting error when you reboot about "invalid raid superblock magic" and you have additional hard drives other than the ones you installed to, check that your hard drive order is correct. During installation, your RAID devices may be hdd, hde and hdf, but during boot they may be hda, hdb and hdc. Adjust your kernel line accordingly. This is what happened to me anyway.

Start arrays read-only

When an md array is started, the superblock will be written, and resync may begin. To start read-only set the kernel module md_mod parameter start_ro. When this is set, new arrays get an 'auto-ro' mode, which disables all internal io (superblock updates, resync, recovery) and is automatically switched to 'rw' when the first write request arrives.

Note: The array can be set to true 'ro' mode using mdadm -r before the first write request, or resync can be started without a write using mdadm -w.

To set the parameter at boot, add md_mod.start_ro=1 to your kernel line.

Or set it at module load time from /etc/modprobe.d/ file or from directly from /sys/.

echo 1 > /sys/module/md_mod/parameters/start_ro

Recovering from a broken or missing drive in the raid

You might get the above mentioned error also when one of the drives breaks for whatever reason. In that case you will have to force the raid to still turn on even with one disk short. Type this (change where needed):

# mdadm --manage /dev/md0 --run

Now you should be able to mount it again with something like this (if you had it in fstab):

# mount /dev/md0

Now the raid should be working again and available to use, however with one disk short! So, to add that one disc partition it the way like described above in Prepare the device. Once that is done you can add the new disk to the raid by doing:

# mdadm --manage --add /dev/md0 /dev/sdd1

If you type:

# cat /proc/mdstat

you probably see that the raid is now active and rebuilding.

You also might want to update your configuration (see: #Update configuration file).


There are several tools for benchmarking a RAID. The most notable improvement is the speed increase when multiple threads are reading from the same RAID volume.

Tiobench specifically benchmarks these performance improvements by measuring fully-threaded I/O on the disk.

Bonnie++ tests database type access to one or more files, and creation, reading, and deleting of small files which can simulate the usage of programs such as Squid, INN, or Maildir format e-mail. The enclosed ZCAV program tests the performance of different zones of a hard drive without writing any data to the disk.

hdparm should NOT be used to benchmark a RAID, because it provides very inconsistent results.

See also


Forum threads

RAID with encryption