tiyn/wiki
1
0
Fork 0
mirror of https://github.com/tiyn/wiki.git synced 2025-04-08 01:27:46 +02:00
Code Issues Projects Releases Wiki Activity
wiki/wiki/linux/mdadm.md

4.5 KiB
Raw Blame History

MDADM

MDADM is a utility to create and manage raid devices. For the rest of this entry n is the number of drives inside a raid device.

Usage

Get Information About a Raid

To get an info for a running raid (assuming it is /dev/md0) run mdadm -D /dev/md0.

Add Disk/Partition to Raid Device

You can add a disk or partition to a raid device by running the following. It is assumed you are adding a partition called /dev/sdd1 but it could also be a whole drive and the mdadm drive is called /dev/md0.

mdadm --add /dev/md0 /dev/sdd1

This will add the disk as spare to the existing raid. To make the disk active the raid needs to be grown as described in the regarding article.

Growing Raid Device

Growing a raid device will increase its number of active drives. Check the number of active drives. Then change and run the following command accordingly.

mdadm --grow --raid-disks=5 /dev/md0

/dev/md0 is the raid device. 5 is the number of disks that should be active. For adding disks view the previous section.

In the following section methods are described that can be tried to affect the performance of the reshaping of a Raid array.

Changing Reshape Speed

Reshaping can take a long time. It can be useful to increase or decrease the speed of it depending on the usage of the raid array while resizing. The following section is based on a guide by nixCraft and addresses the temporary change of speed of the reshape process of a raid array.

There are two values that can be temporarily changed to limit the reshape speed. Those are dev.raid.speed_limit_min which defaults to 1000 and dev.raid.speed_limit_max which defaults to 10000.

sysctl -w dev.raid.speed_limit_min=100000
sysctl -w dev.raid.speed_limit_max=500000

Another value to change is the read-ahead. The current value can be displayed with the following command.

blockdev --getra /dev/md0

And it can be set with the following command. This commands sets the read-ahead to 32MB.

blockdev --setra 65536 /dev/md0

For Raid 5 the value stripe_cache_size can be important too. It can be changed with the following command which sets it to 32 MiB.

echo 32768 > /sys/block/md3/md/stripe_cache_size

However make sure not to set it too large, as this can result in a "out of memory" condition. The memory consumed can be calculated with the following formula.

memory_consumed = system_page_size * nr_disks * stripe_cache_size

Raid 1

Raid 1 creates a mirror with even amount of drives. For n=2 raid 5 and raid 1 are basically the same. The space efficiency is 1/n and the fault tolerance is n-1 drive failure. The read perfromance is n and the write performance 1.

Create Raid 1 Device

You can create a Raid 1 device with mdadm --create --verbose /dev/md0 --level=1 --raid-devices=2 /dev/sda /dev/sdb where sda and sdb are the disks you want to include in your raid. You can increase the number of raid devices, but they have to be at least 2. If it is an uneven amount of disks in the raid, a disk will act as a spare disk.

Convert Raid 1 to Raid 5

Assuming the raid 1 device is called /dev/md0. All other drives are part of the md0 raid device. Note that mostly raid 1 devices consisting of 2 drives should be converted to raid 5.

  • Remove all drives but 2 (if there are more drives than that) by running mdadm /dev/md0 --fail /dev/sda1 and mdadm /dev/md0 --remove /dev/sda1 where sda1 is the drive to remove
  • Make sure your raid 1 array has only 2 active drives by running mdadm --grow /dev/md0 -n 2
  • Now convert your raid 1 to a raid 5 device with mdadm --grow /dev/md0 -l5
  • Add the disks you removed back to the raid with mdadm /dev/md0 --add /dev/sda1
  • Finally grow the active drive number to your needs (in this example 4) mdadm --grow /dev/md0 -n4
  • MDADM now reshapes the raid. You can monitor it by running watch cat /proc/mdstat

Raid 5

Raid 5 creates a raid device with distributed parity and is set to have at least 3 drives. The space efficiency is 1-(1/n) and the fault tolerance is 1 drive failure. The read perfromance is n and the write performance 1/4 for single sector and n-1 for full stripe.

In the special case of 2 drives in a raid 5 it is functionally the same as raid 1.