Data Protection: RAID

Chapter Objective
After completing this chapter, you will be able to: Describe what is RAID and the needs it addresses Describe the concepts upon which RAID is built Define and compare RAID levels Recommend the use of the common RAID levels based on performance and availability considerations Explain factors impacting disk drive performance

Data Protection: RAID

Why RAID
Performance limitation of disk drive An individual drive has a certain life expectancy Measured in MTBF Example If the MTBF of a drive is 750,000 hours, and there are 1000 drives in the array, then the MTBF of the array becomes 750,000 /1000, or 750 hours RAID was introduced to mitigate this problem RAID provides: Increase capacity Higher availability Increased performance

Data Protection: RAID

RAID Array Components
Physical Array

Logical Array

RAID Controller

Hard Disks

Host
RAID Array

Data Protection: RAID

RAID Implementations
Hardware (usually a specialized disk controller card) Controls all drives attached to it Array(s) appear to host operating system as a regular disk drive Provided with administrative software Software Runs as part of the operating system Performance is dependent on CPU workload Does not support all RAID levels

Data Protection: RAID

RAID Levels
0 Striped array with no fault tolerance 1 Disk mirroring Nested RAID (i.e., 1 + 0, 0 + 1, etc.) 3 Parallel access array with dedicated parity disk 4 Striped array with independent disks and a dedicated parity disk 5 Striped array with independent disks and distributed parity 6 Striped array with independent disks and dual distributed parity

Data Protection: RAID

Data Organization: Striping
Stripe

Strip

Stripe=192KB

Strip 1=64KB

Strip 2=64KB

Strip 3=64KB

Stripe 1 Stripe 2
Strips

Data Protection: RAID

RAID 0
0

1 5 9 RAID Controller

2 6 10
3 7 11

Host

Data Protection: RAID

RAID 1

Block 0 1

RAID Block 0 1 Controller

Host

Data Protection: RAID

Nested RAID 0+1 (Mirrored Stripe)
RAID 1

Block 1 Block 4

Block 3 0

RAID Block 4 2 1 Block 5 Controller

Block 2

RAID 0

Host

Block 5

Data Protection: RAID

Nested RAID 0+1 (Striping and Mirroring)
RAID 1

Block 1 Block 4 RAID Controller Block 2

Host RAID 0

Block 1 Block 4

Block 2 Block 5

Block 5

Data Protection: RAID

Nested RAID 1+0 (Mirroring and Striping)
RAID 0

1 Block 2
Block 4 5 RAID Controller 1 Block 2
Host RAID 1

Block 3 0

Block 4 5

Data Protection: RAID

Nested RAID 1+0 (Mirroring and Striping)
RAID 0

Block 1 Block 4 RAID Controller Block 1

Host RAID 1

Block 2 Block 5

Block 2 Block 5

Block 4

Data Protection: RAID

RAID Redundancy: Parity
0

4
1 5 9

RAID Controller

1 ?

Host

The middle drive fails: Parity calculation 4 + 6 + 1 + 7 = 18

3 7 11

4 + 6 + ? + 7 = 18 ? = 18 4 6 7 ?=1

0123 4567 18

Parity Disk

Data Protection: RAID

RAID 3

Block

RAID Block 0 Controller Block 1 Parity Generated Block 2

Block 3 P0123

Host

Data Protection: RAID

RAID 4
Block 0 Block 4 Block 1 Block 5 Parity RAID Block 0 Generated Controller P0123 Block 3
Host

Block 0

Block 2
Block 6

Block 7 P0123 P4567

Data Protection: RAID

RAID 5
Block 0 Block 4 Block 1 Block 5 Parity RAID Block 4 0 Generated Controller P4 05 16 27 3 Block 3
Host

Block 0 4

Block 2
Block 6

P4567 P0123 Block 7

Data Protection: RAID

RAID 6 Dual Parity RAID
Two disk failures in a RAID set leads to data unavailability and data loss in single-parity schemes, such as RAID-3, 4, and 5 Increasing number of drives in an array and increasing drive capacity leads to a higher probability of two disks failing in a RAID set RAID-6 protects against two disk failures by maintaining two parities Horizontal parity which is the same as RAID-5 parity Diagonal parity is calculated by taking diagonal sets of data blocks from the RAID set members Even-Odd, and Reed-Solomon are two commonly used algorithms for calculating parity in RAID-6

Data Protection: RAID

RAID Comparison
RAID Min Disks Storage Efficiency % Cost Read Performance
Very good for both random and sequential read

Write Performance

100

Low

Very good

50 (n-1)*100/n where n= number of disks (n-1)*100/n where n= number of disks (n-2)*100/n where n= number of disks

High

Good Better than a single disk

Good Slower than a single disk, as every write must be committed to two disks Poor to fair for small random writes Good for large, sequential writes Fair for random write Slower due to parity overhead Fair to good for sequential writes Good for small, random writes (has write penalty)

Moderate

Good for random reads and very good for sequential reads Very good for random reads Good for sequential reads Very good for random reads Good for sequential reads

Moderate

Moderate but more than RAID 5

1+0 and 0+1

50

High

Very good

Good

Data Protection: RAID

RAID Impacts on Performance
RAID Controller

Ep new

Ep old

2 XOR

E4 old

E4 new

P0

D1

D2

D3

D4

Small (less than element size) write on RAID 5

Ep = E1 + E2 + E3 + E4 (XOR operations) If parity is valid, then: Ep new = Ep old E4 old + E4 new (XOR operations)
2 disk reads and 2 disk writes
Reading, calculating and writing parity segment introduces penalty to every write operation Parity RAID penalty manifests due to slower cache flushes Increased load in writes can cause contention and can cause slower read response times

Parity Vs Mirroring

Data Protection: RAID

RAID Penalty Exercise
Total IOPS at peak workload is 1200 Read/Write ratio 2:1 Calculate IOPS requirement at peak activity for RAID 1/0 RAID 5

Data Protection: RAID

Hot Spares

RAID Controller

Data Protection: RAID

Summary
Key points covered in this chapter: What RAID is and the needs it addresses The concepts upon which RAID is built Some commonly implemented RAID levels