RAID

• Hard dirives are relatively fast, persistent storage

Shortcomings

• How to cope with disk failure?
  • Mechanical parts break over time
  • Sector may become silently corrupted
• Capacity is limited
  • Managing files across multiple physical devices is cumbersome
    -> parallel

---

RAID

Redundant Array of Inexpensive(=Independent) Disks

• Use multiple disks to create the illusion of a large, faster, more reliable disk
• RAID looks like a single disk
• Internally, RAID is a complex computer system
  • Disks managed by a dedicated CPU + software
  • RAM and non-volatile memory
  • Many different RAID levels

---

RAID 0: Stripping

present an array of disks as a single large disk

여러 드라이브로 분산되지만, 사용자는 마치 하나의 큰 디스크를 사용하는 것처럼 느낌

datablock -> 512 byte(sector)

• Sequential accesses spread accros all drives
• Random accesses are naturally spread over all drives

Addressing Blocks

• Disk = logical block number % number of disks
• Offset = logical block number / number of disks
• E.g. read block 11
  • 11 % 4 = Disk 3
  • 11 / 4 = Physical Block 2(starting from 0)
    

Chuck sizing

• Chuck size impacts array performance
• Smaller chucks -> greater parallelism
• Big chunks -> reduced seek times
  • Typical array use 64KB chucks

seektime 때문에 chuck size가 클 때가 병렬성이 감소하더라도 더 빠를 수 있음

Measuring RAID Performance

1. Sequential access time S

• 10MB transfer
• $S$ = transfer size / time to access
  

  sequential은 처음에만 seektime 존재

2. Random access time R

• 10KB transfer
• $R$ = transfer size / time to access
  
• On each access, need seek time and rotational delay

Analsys of RAID 0

• Capacity : $N$ (Number of Disks)
• Reliability : 0
• Sequential read and write : $N∗S$
• Random read and write : $N∗R$

---

RAID 1: Mirroring

RAID 0 offers high performance, but zero error recovery
-> Make two copies of all data

RAID 0+1 and 1+0

Analysis of RAID 1

• Capacity : $N/2$ - mirroring으로 half
• Reliability : 1 drive can fail
  • Lucky -> $N/2$ drives can fail without data loss

거의 RAID 0 성능의 절반이지만, Reliability good but expensive

• Sequential write : $(N/2)/S$
  • Two copies of all data, thus half throughput
• Sequential read : $(N/2)/S$
• Random write : $(N/2)/R$
  • Two copies of all data, thus half throughput
    -> 둘 다 써줘야 하니까 /2
• Random read: $(N)/R$
  • Best scenario for RAID 1
  • Reads can parallelize across all disks

Consistent update problem

• Mirrored writes should be atomic
  • All copies are written or none are written
• Difficult to guarantee (e.g. power failure)
• Many RAID controllers include a write-ahead log
  • Battery backed, non-volatile storage
    

disk0에 쓰다가 power 나가면 disk2와 다른 데이터가 됨 -> cache

---

RAID 4: Parity Drive

Disk N only stores parity information for the other N-1 disks

Updating Parity on Write

1. Additive parity

• Big overhead
• Write Disk 4를 원했으나, Read 3번(0, 1, 3), Write 2번(2,4)
  

2. Subtract parity

• Write 2, 4
• Disk 2 : 1 -> 0
  

Analysis of RAID 4

• Capacity : $N-1$
  • Space on the parity drive is lost
• Reliability : 1 drive can fail
• Sequential Read and Write : $(N-1)*S$
  • Parallelization across all non-parity blocks
  • stripe 단위이므로 한 parity bit만 update
• Random Read : $(N-1)*R$
  • Reads parallelize over all but the parity drive
• Random Write : $R/2$
  • Writes serialize due to the parity drive
  • Each write requires 1 read and 1 write of the parity drive
  • All writes must update the parity drive, causing serialization

---

RAID 5: Rotating Parity

Parity blocks are spread over all N disks

Analysis of RAID 5

분산되어 있지만 한 군데로 모으면 same as RAID 4

• Capacity : $N-1$
  • Space on the parity drive is lost
• Reliability : 1 drive can fail
• Sequential Read and Write : $(N-1)*S$
  • Parallelization across all non-parity blocks
• Random Read :$N*R$
  • Reads parallelize over all drives
• Random Write :$N/4*R$
  • Unlike RAID 4, writes parallelize over all drives
  • Each write requires 2 reads and 2 write, hence N / 4
    

---

RAID 6

Two parity blocks per stripe

• Any two drives can fail
• Capacity : $N−2$
• No overhead on read, significant overhead on write
• Implemented using Reed-Solomon codes

---

Choosing a RAID Level

• Best performance
  • RAID 0
• Greatest error recovery
  • RAID 1 or RAID 6
• Balance
  • RAID 5

---

Other Considerations

Many RAID systems include a hot spare

• Unused disk installed in the system -> 여분의 disk, 무중단 system
• If a drive fails, the array is immediately rebuilt using the hot spare

RAID can be implemented in HW or SW

• Hardware
  • Faster and more reliable
  • Migrating to a different hardware controller almost never works
• Software
  • Simpler to migrate and cheaper
  • Worse performance and weaker reliability
    • Due to the consisten update problem -> 비휘발성 메모리를 쓰지 못해서