Software RAID in Linux Workstations

Some software RAID schemes offer data protection through mirroring (copying the data to multiple

disks in case one fails) or parity data (checksums that allow error detection and limited rebuilding of

data in case of a failure), but it should be noted that all software RAID solutions on workstations will

require the shutdown of the system so that the failed drive can be replaced before redundancy can be

restored. Currently only external hardware RAID allows online, invisible failure recovery through a

hot-swap, however, the replacement for failed drives in software RAID requires only a minimum

amount of work. Hardware RAID is not supported by HP Linux workstations.

Performance and Bottlenecks

Disk I/O bandwidth is typically limited by the system’s bus speeds, the disk controller, and the disks

themselves. Each generation of disks tends to get faster and newer rotation speeds premiere from time

to time. The balance of these hardware limiters, as affected by the software configuration, determines

where the real bottleneck is in the system.

Several RAID levels offer improved performance relative to a standalone disk. If your disk throughput

is throttled by a single disk controller, there is probably little you can do with RAID to improve the

performance without adding another controller. On the other hand, if the raw disk performance is the

bottleneck, a tuned software RAID solution can dramatically improve the throughput. In fact, the

slower the disk is relative to the rest of the system, the better RAID performance will scale, since the

slowest piece of the performance pipeline is being directly addressed by moving to RAID.

RAID Levels

RAID-Linear: Concatenating Disks

RAID-linear can only generously be considered to be in the RAID family. It is generally used as a

method of making two smaller disks look like a larger one by “appending” one to the other. For

instance, two 40GB disks configured with RAID-linear look like one 80GB disk. RAID-linear offers no

practical performance or reliability gains over standard separate disks. Spare disks are not supported

in RAID-linear; if one disk fails, the RAID array fails as well. RAID-linear is not supported by HP Linux

workstation.

RAID-0: Striping

RAID-0, or striping, means that the data is spread out in small chunks over multiple disks so that a

read or write operation can fully utilize all the disk and disk controller bandwidth available by having

each disk and/or disk controller share the load of any particular operation. The drives in the array

are treated as one large address space and successive data blocks are typically written to different

drives in the array. RAID-0 arrays are comprised of a minimum of two disk drives. In order to optimize

read and write speeds, partitions in RAID-0 arrays should be located on different physical hard disks.

RAID-0 was not included in the original RAID proposal as it does not technically meet the redundancy

requirement, nor does it offer any error protection. Software RAID-0 is supported by HP Linux

workstations.