In addition to creating RAID arrays, mdadm can also be used to take advantage of hardware supporting more than one I/O path to individual SCSI LUNs (disk drives). The goal of multipath storage is continued data availability in the event of hardware failure or individual path saturation. Because this configuration contains multiple paths (each acting as an independent virtual controller) accessing a common SCSI LUN (disk drive), the Linux kernel detects each shared drive once "through" each path. In other words, the SCSI LUN (disk drive) known as /dev/sda may also be accessible as /dev/sdb, /dev/sdc, and so on, depending on the specific configuration.

To provide a single device that can remain accessible if an I/O path fails or becomes saturated, mdadm includes an additional parameter to its ––level option. This parameter — multipath — directs the md layer in the Linux kernel to re-route I/O requests from one pathway to another in the event of an I/O path failure.

To create a multipath device, edit the /etc/mdadm.conf file to define values for the DEVICE and ARRAY lines that reflect your hardware configuration.

The command used for the creation of a multipath device is similar to that used to create a RAID device; the difference is the replacement of a RAID level parameter with the multipath parameter:

mdadm -C /dev/md0 --level=multipath --raid-devices=4 /dev/sda1 /dev/sdb1 /dev/sdc1 /dev/sdd1
Continue creating array? yes
mdadm: array /dev/md0 started.

Due to the length of the mdadm command line, it has been broken into two lines.

In this example, the hardware consists of one SCSI LUN presented as four separate SCSI devices, each accessing the same storage by a different pathway. Once the multipath device /dev/md0 is created, all I/O operations referencing /dev/md0 are directed to /dev/sda1, /dev/sdb1, /dev/sdc1, or /dev/sdd1 (depending on which path is currently active and operational).

The configuration of /dev/md0 can be examined more closely using the command mdadm --detail /dev/md0 to verify that it is, in fact, a multipath device:

/dev/md0:
Version : 00.90.00
Creation Time : Tue Mar  2 10:56:37 2004
Raid Level : multipath
Array Size : 3905408 (3.72 GiB 3.100 GB)
Raid Devices : 1
Total Devices : 4
Preferred Minor : 0
Persistence : Superblock is persistent

Update Time : Tue Mar  2 10:56:37 2004
State : dirty, no-errors
Active Devices : 1
Working Devices : 4
Failed Devices : 0
Spare Devices : 3

    Number   Major   Minor   RaidDevice State
       0       8       49        0      active sync   /dev/sdd1
       1       8       17        1      spare   /dev/sdb1
       2       8       33        2      spare   /dev/sdc1
       3       8        1        3      spare   /dev/sda1
           UUID : 4b564608:fa01c716:550bd8ff:735d92dc
         Events : 0.1

Another feature of mdadm is the ability to force a device (be it a member of a RAID array or a path in a multipath configuration) to be removed from an operating configuration. In the following example, /dev/sda1 is flagged as being faulty, is then removed, and finally is added back into the configuration. For a multipath configuration, these actions would not affect any I/O activity taking place at the time:

# mdadm /dev/md0 -f /dev/sda1
mdadm: set /dev/sda1 faulty in /dev/md0
# mdadm /dev/md0 -r /dev/sda1
mdadm: hot removed /dev/sda1
# mdadm /dev/md0 -a /dev/sda1
mdadm: hot added /dev/sda1
#