NotesAtHKU

Hard disk drives

Components

Platter: A disk that stores data magnetically.
Tracle: A ring on a platter that stores data.
Sector: A segment of a track, usually 512 bytes.
Cylinder: A set of tracks on different platters that are aligned vertically.
Read/write head: A device that reads and writes data on a platter.

Sector format

30 sectors, 600 bytes each, 512 bytes for data, 88 bytes for control information
Sector fields:
- Gap 1 (17b): Separates sectors
- ID Field (7b): Contains Synch (1b), Track, head, sector (4b), CRC (2b)
- Gap 2 (41b): Separates ID and data field
- Data Field (515b): Contains Synch (1b), Data (512b), CRC (2b)
- Gap 3 (40b): Separates data and ID field

Disk layout methods

There are two methods for laying out data on a disk:

Constant Angular Velocity (CAV): The disk rotates at a constant speed, and the data is read at a constant rate. Easier read/write, but density of data decreases from inner track to outer tracks, wasting space.
Multiple Zone Recording (MZR): Divides disk into zones, with each zone having a different number of sectors per track. Data density is nearly constant and allows maximized storage capacity.

Disk Access Time

Average seek time $T_\text{seek}$ : Time taken to move the read/write head to the desired track.
Average rotational latency $T_\text{latency}$ : Time required for sector to rotate under the head
Data transfer time $T_\text{transfer}$ : Time taken to read or write the data once the head is in position.
Time for one rotation $T_\text{rotation}$
TIme to rotate one sector $T_\text{sector}$
Number of sectors per track $n_\text{sectors}$

Formulas in seconds:

\begin{aligned} & \frac{1}{rps} = \frac{60}{RPM} \\ & T_\text{rotation} & \frac{1}{rps} \\ & T_\text{latency} & \frac12 \times T_\text{rotation} \\ & T_\text{sector} & \frac{T_\text{rotation}}{n_\text{sectors}} \\ & T_\text{transfer} & \frac{\text{bytes}}{\text{bytes per track}} \times \frac{1}{rps} \\ \end{aligned}

Time to access $n$ consecutive sectors ignoring transfer speed is given by $T_\text{seek} + T_\text{latency} + n \times T_\text{sector}$

Time to access $n$ non-consecutive sectors ignoring transfer speed is given by $(T_\text{seek} + T_\text{latency} + T_\text{sector}) \times n$

Solid State Drives (SSD)

SSD

Non-volatile storage devices
Limited write cycles
Faster I/O, access time than HDDs
No moving parts: Lower power consumption, less heat, no noise, high durability

Redundant Array of Independent Disks (RAID)

The goal of RAID is to improve performance and reliability of data storage by using multiple disks.

RAID Level	Data Layout / Parity	Fault Tolerance	Advantages	Disadvantages
0	Data is striped (distributed in a round-robin fashion) across disks; no parity	None – one disk failure causes total data loss	Excellent I/O performance; simple design	No redundancy – loss of any disk results in loss of the entire array
1	Data is mirrored on two disks	Can tolerate disk failure as long as one copy remains	100% redundancy; potentially reduced seek time during read operations	Requires double the physical writes; highest storage overhead
2	Data is striped at the bit level with dedicated disks for Error Correction Codes (Hamming codes)	Error correction possible, but requires extra disks	High data transfer rate (especially with smaller strip sizes); simpler ECC controller design compared to later RAID levels	Expensive; high ECC disk-to-data disk ratio; not used commercially
3	Data is striped at the bit level; one dedicated disk stores parity bits	Recovers from a single disk failure	Very high read/write speeds; efficient use of ECC for fault recovery	Dedicated parity disk can become a bottleneck
4	Data is striped at the block level with a dedicated parity disk	Recovers from a single disk failure	Allows individual spindle control; similar benefits as RAID 3	Write penalty (requiring extra read/write cycles to update parity); uncommon
5	Data and parity are distributed across all disks; parity is calculated on a block basis	Can endure one disk failure	High I/O performance; distributed parity avoids single-disk bottlenecks; widely used	More complex parity recalculation; rebuild process can be challenging
6	Similar to RAID 5 but with dual distributed parity (using two different parity methods)	Tolerates up to two disk failures	Highest fault tolerance; enhanced data reliability	Higher controller complexity and parity computation overhead

External storage