RAID

Stripe Width

Refers to the maximum number of drives used for parallelization of the IO requests.

Stripe Size

The standard storage segment size on each drive in the array. The stripe size is used to tune the IO speed based on the actual usage. • Let stripe-width be 2 and stripe-size be 16KB, show how to distribute these files: • File-1 = 4 KB • File-2 = 20 KB • File-3 = 100 KB • File-1 is on D1 first block. • File-2 is on D1 and D2: • 12KB on d1 and 8KB on d2 • File-3 is on D1 and D2 • 8KB on first block of D2 • Five 16KB blocks distributed across D1 and D2 • 12KB on D2.

RAID 1

• AKA, Mirroring A second drive is used to duplicate everything on the main drive. Achieves redundancy • Easy to implement, but doubles the cost of disk storage. • Writes are slightly slower due to the overhead of managing the copy. • Can improve overall read performance. There are three different approaches to RAID-1 reads: • The controller can handle two separate requests simultaneously. • The controller can stripe across the two drives reading alternate blocks or bytes. • Neither of the above, simply read as if the array is a single HDD

Multiple RAID Levels

• Beyond these single-level RAID designs, a number of multiple RAID levels have been defined, which use two or more of the single RAID levels in combination to create new array types with new capabilities (and limitations). • Many have largely disappeared from the market as experience over time has shown them to be inferior to other levels without advantages to compensate.

RAID 2

• Bits instead of bytes are interleaved across multiple drives. • Here, along with the 64 data bits, an additional 8 bits of ECC (Error Correction Code) codes are stored. • Not implemented.

RAID 1+0

• Combines RAID 0 and RAID 1 • Striping & mirroring • RAID 0 Performance. • RAID 1 Reliability. • Doubles the cost of storage.

RAID 0+1

• Combines RAID-0 and 1 • Lower fault tolerance compared to RAID 1+0

RAID 5

• Combines striping and distributed parity. Ideal for small IO requests. • Multiple disks are configured so data is striped across them. • There is a parity region on each disk used to rebuild if a disk in the array should fail. • Parity bit computation and updates slows write performance. • Read performance can be excellent. • Stripe width = n, where n equals the number of disks in the array. Writes perform best with n-1 stripe width. • Fault tolerance is 1 HDD. • Stripe size is at the block level. • One of the most common implementations of RAID at the enterprise level.

Warm Swap

• Computing system access is suspended, but not powered-down, while the file system is repaired and rebuilt.

RAID 0

• Data is spread across multiple disks • Enhances speed for medium to large size IO (read and write) requests. Small IO requests do not benefit. • Degrades reliability by a factor of n, where n is the number of disks in the array. MTBF degrades as a function of n.

Common Raid Implementations

• RAID 0 • RAID 1 • RAID 2 • RAID 3 • RAID 4 • RAID 5 • RAID 6 • RAID 1+0 • RAID 0+1

RAID

• Redundant Array of Independent Disks. • Use multiple drives to achieve: • Improved performance • Improved reliability via redundancy.

RAID 4

• Similar to RAID 3. • Same stripe width as RAID-3, n-1 • One dedicated parity drive • Stripe size is at the block level. • The parity drive becomes a bottleneck for writes. Performs best with long IO ops. • Rare implementation

RAID 6

• Similar to RAID 5. • More reliable compared to RAID 5, tolerates two drive failures • Fault tolerance is 2 HDDs. • Greater overhead slows writes due to 2-parity calculations and parity updates. • Best suited for low cost large storage arrays

RAID Software vs Hardware Implementation

• Software vs Hardware implementation • Enterprise level systems almost always implement RAID using a hardware controller, rather than by software implementation. • The purpose of RAID is to improve performance. Software implementations are not as efficient.

RAID 3

• Striped at the byte level, implements parity. • Data is striped across two or more HDDs. • All drives operate in parallel • Parity bits are stored on a dedicated drive. • Ideal for large IO requests. • Parity is achieved by the XOR Boolean operation. • For an n-disk RAID-3 array, the stripe width is n-1. • Fault tolerance is 1 failed HDD. • Performance will be slow if there is a disk failure due to parity calculations. • Byte level striping • RAID-3 requires synchronized spinning of the array. Difficult to implement. RARE. • Performs best with long reads & writes. Performs slowly with short IO ops.

Cold Swap

• System must be shutdown to replace a failed drive. • File system is not available while being restored.

Hot Swap

• The file system remains available while being repaired and rebuilt. However access has a performance hit. • Performance of reads requires reading parity and doing a parity-restore calculation.

Hot Standby

• This facility is provided by the RAID controller card. • An additional disk is connected to the array but is not used for IO access unless there is a failure. • The data and parity information is recreated onto the "spare" drive if needed.