24. Raid

Info

This is a summary of Chapter 38 from the book Operating System: Three Easy Pieces by Remzi H. Arpaci-Dusseau and Andrea C. Arpaci-Dusseau. This chapter focuses on Redundant Arrays of Inexpensive Disks (RAID), explaining different RAID levels, their advantages, trade-offs, and performance characteristics.

1. Introduction

Disks are often slow, limited in capacity, and vulnerable to failure. RAID systems address these issues by combining multiple disks to improve:

1. Performance – Faster I/O operations via parallelism.
2. Capacity – Larger storage by combining multiple disks.
3. Reliability – Fault tolerance by introducing redundancy.

Key Questions:

• How can RAID enhance disk storage?
• What trade-offs exist between different RAID levels?

---

2. RAID Overview

RAID presents a transparent interface to the operating system, appearing as a single large disk. Internally, it consists of:

• Multiple disks working together.
• A RAID controller managing disk interactions.
• Memory (DRAM/NVRAM) for caching and buffering.
• Specialized hardware/software for redundancy management.

RAID Benefits:

• Improves performance through parallel I/O.
• Increases storage capacity by aggregating disks.
• Enhances reliability via redundancy mechanisms.

---

3. RAID Fault Model

RAID assumes a fail-stop model:

• Disks are either fully operational or completely failed.
• Failures are immediately detectable by the RAID controller.
• More complex failures (e.g., silent corruption, latent sector errors) are not initially considered but are relevant in real-world scenarios.

---

4. RAID Performance Metrics

RAID levels are evaluated based on:

1. Capacity – Usable storage after redundancy.
2. Reliability – Fault tolerance (number of disks that can fail).
3. Performance – Impact on read/write speeds.

---

5. RAID Levels and Analysis

RAID 0 (Striping)

• No redundancy (not technically a RAID).
• Distributes data across disks to maximize performance.
• Capacity: $(N\times B)$
• Reliability: 0 (failure of one disk = data loss).
• Performance:
  • Sequential Read/Write: $(N\times S)$ (uses all disks).
  • Random Read/Write: $(N\times R)$.

---

RAID 1 (Mirroring)

• Each disk has an exact copy on another disk.
• Tolerates single disk failure (or more, if lucky).
• Capacity: $(N\times R)/2$.
• Reliability: 1 disk (or more, depending on failures).
• Performance:
  • Reads: $(N\times R)$ (can use either copy).
  • Writes: $(N/2\times R)$ (must write to both copies).

---

RAID 4 (Parity-Based)

• Uses one disk for parity calculations.
• Recovers lost data using XOR calculations.
• Capacity: $((N-1)\times B)$.
• Reliability: 1 disk failure tolerance.
• Performance:
  • Sequential Read/Write: $((N-1)\times S)$.
  • Random Read: $((N-1)\times R)$.
  • Random Write: Severely bottlenecked by parity disk.

🔴 Small-Write Problem: Every write requires updating the parity disk, causing a bottleneck.

---

RAID 5 (Rotating Parity)

• Same as RAID 4, but parity is distributed across disks.
• Reduces the parity disk bottleneck, allowing better performance.
• Capacity: $((N-1)\times B)$.
• Reliability: 1 disk failure tolerance.
• Performance:
  • Sequential Read/Write: $((N-1)\times S)$.
  • Random Read: $(N\times R)$.
  • Random Write: $(N/4\times R)$ (better than RAID 4).

---

6. Summary Table

RAID	Capacity	Reliability	Seq. Read	Seq. Write	Rand. Read	Rand. Write
RAID 0	$(N\times B)$	0	$(N\times S)$	$(N\times S)$	$(N\times R)$	$(N\times R)$
RAID 1	$((N\times B)/2)$	1 (or more if lucky)	$(N\times R)$	$(N/2\times S)$	$(N\times R)$	$(N/2\times R)$
RAID 4	$((N-1)\times B)$	1 disk	$((N-1)\times S)$	$((N-1)\times S)$	$((N-1)\times R)$	$(R/2)$ (Bottlenecked)
RAID 5	$((N-1)\times B)$	1 disk	$((N-1)\times S)$	$((N-1)\times S)$	$(N\times R)$	$(N/4\times R)$

---

7. Conclusion

• RAID 0: Best performance, no reliability.
• RAID 1: Best reliability, expensive (requires 2x storage).
• RAID 4: Efficient storage, poor write performance.
• RAID 5: Good balance of reliability and storage efficiency.

Choosing the right RAID level depends on whether performance, reliability, or storage efficiency is the priority.