Fun with RAID

Jozef

RAID

• apt-get install mdadm
• # dpkg-reconfigure mdadm

RAID

Redundant Array of Independent/Inexpensive Disks

• RAID0 - striping
  n*SIZE
• RAID1 - mirror
  SIZE
• RAID5 - striped set with distributed parity
  (n-1)*SIZE
• RAID6 - striped set with dual parity
  (n-2)*SIZE

software raid

• more flexible
• can be reconfigured "on the fly"
• ex. pvmove data from RAID1 blockdevice,
  destroy the RAID1 and use the disks to construct RAID0, vgextend volume gropu with twice the size
• ex.2 replacing disks for bigger ones
• cat /proc/mdstat

server with 2 disks

• /dev/hda1 + /dev/hdb1 = RAID1 => /boot
• /dev/hda2 + /dev/hdb2 = RAID1 => /
• /dev/hda3 + /dev/hdb3 = RAID0 => /data
• redundacy for /boot and /, speed for data
• in case of 1 disk failure system will survive
  no need for reinstalation, data are lost but recovered from backup

server with 3 disks

• /dev/hda1 + /dev/hdb1 = RAID1 => /boot
  /dev/hdc1 spare
• /dev/hda2 + /dev/hdb2 = RAID1 => /
  /dev/hdc2 spare
• /dev/hda3 + /dev/hdb3 + /dev/hdc3 = RAID5 => /data
• redundacy for /boot and /, speed+redundancy for data
• server runs with no impact in case of 1 disk failure
  if two disks fail (not at the same time) system is ok but /data lost

server with 8 disks

• /dev/sda + /dev/sdb = RAID1
  /dev/sdc spare (3 disks)
• /dev/sdd + /dev/sde + /dev/sdf + /dev/sdg = RAID5 => /data
  /dev/sdh spare (5 disks)

destroy our lvm setup...

• lvremove vg00/test1
• vgremove vg00
• pvremove /dev/hdb # /dev/hdd (wiped with badblocks)

create raid0

• mdadm --create /dev/md0 --level=0 --raid-devices=2 /dev/hd[bd]
• cat /proc/mdstat
• fdisk -l /dev/md0
• mkfs.ext3 /dev/md0
• mount /dev/md0 /mnt
• umount /mnt
• mdadm --stop /dev/md0 --zero-superblock /dev/hd[bd]

create raid1

• mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/hd[bd] --spare-devices=1 /dev/hda5
• fdisk -l /dev/md0
• cat /proc/mdstat
• mkfs.ext3 /dev/md0
• mount /dev/md0 /mnt
• mdadm --manage /dev/md0 --fail /dev/hdb --fail /dev/hda5
• mdadm --remove /dev/md0 /dev/hdb /dev/hda5
• dd if=/dev/zero of=/dev/hdb
• dd if=/dev/zero of=/dev/hdd
• mdadm --add /dev/md0 /dev/hdb /dev/hda5
• umount /mnt
• mdadm --stop /dev/md0 --zero-superblock /dev/hd[bd] /dev/hda5

notes

• /dev/hda5 1310 1374 522081 fd Linux raid autodetect
• take care when buying different hdd-s
  80GB is not always 80*1024*1024*1024B
  80GB is sametimes 80*1000*1000*1000B
  The difference is 1920MB or 2_013_265_920B...

And that's it...

Questions?