A traditional Hard Disk Drive (HDD) consists of:
Total disk access time = Seek Time + Rotational Latency + Transfer Time
Seek time dominates - disk scheduling algorithms aim to minimize total seek time.
| Component | Description | Typical Value |
|---|---|---|
| Seek Time | Time to move the read/write head to the correct track | 3-15 ms |
| Rotational Latency | Time for the desired sector to rotate under the head | 0-8 ms (avg ~4 ms at 7200 RPM) |
| Transfer Time | Time to read/write the actual data | Very small (microseconds) |
| Feature | HDD | SSD |
|---|---|---|
| Technology | Magnetic platters, mechanical arm | NAND flash memory (no moving parts) |
| Speed (sequential) | 100-200 MB/s | 500 MB/s - 7 GB/s (NVMe) |
| Random access | Slow (seek time) | Very fast (microseconds) |
| Durability | Vulnerable to physical shock | More durable (no moving parts) |
| Noise | Audible (spinning, seeking) | Silent |
| Power consumption | Higher | Lower |
| Cost per GB | Lower (~$0.02/GB) | Higher (~$0.08/GB) |
| Lifespan | 3-5 years (mechanical wear) | Limited write cycles (TBW) |
RAID (Redundant Array of Independent Disks) combines multiple physical disks into a logical unit for improved performance, redundancy, or both.
| RAID Level | Description | Min Disks | Fault Tolerance | Use Case |
|---|---|---|---|---|
| RAID 0 (Striping) | Data split across disks for speed. No redundancy. | 2 | None - any disk failure = total data loss | High performance, non-critical data |
| RAID 1 (Mirroring) | Exact copy of data on two disks. | 2 | 1 disk failure | OS drives, critical data |
| RAID 5 (Striping + Parity) | Data and parity striped across all disks. Parity allows recovery. | 3 | 1 disk failure | File servers, general storage |
| RAID 6 (Double Parity) | Like RAID 5 but with two parity blocks. | 4 | 2 disk failures | Large arrays, high availability |
| RAID 10 (1+0) | Mirrored pairs that are then striped. Best of RAID 1 and RAID 0. | 4 | 1 disk per mirrored pair | Databases, high I/O workloads |
The storage hierarchy organizes storage by speed, cost, and capacity. Faster storage is more expensive and has less capacity:
Principle of Locality: Programs tend to access a small portion of their address space at any given time (temporal and spatial locality). This is why caching is effective - frequently accessed data is kept in faster storage levels.
| Level | Type | Speed | Capacity | Volatile? |
|---|---|---|---|---|
| 1 | CPU Registers | ~1 ns | Bytes | Yes |
| 2 | L1/L2/L3 Cache | 1-10 ns | KB-MB | Yes |
| 3 | Main Memory (RAM) | ~100 ns | GB | Yes |
| 4 | SSD (NVMe) | ~100 us | TB | No |
| 5 | HDD | ~10 ms | TB | No |
| 6 | Optical / Tape | Seconds | TB-PB | No |
SSDs remove seek and rotational delay, but controller queues, flash translation, garbage collection, endurance, and thermal limits still affect latency. TRIM or discard tells the device which blocks no longer hold live data, while overprovisioning and write patterns influence sustained performance.
RAID protects against selected device failures; it does not protect against deletion, corruption, malware, controller mistakes, or site loss. Define the failure domain, monitor rebuild risk, keep independent backups, and test restoration. Compare tail latency and degraded-mode behavior, not only peak sequential throughput.
Event 1: a process requests a resource related to Storage
Event 2: the kernel checks permissions, availability, and current state
Event 3: the scheduler or manager decides whether to run, wait, block, or fail
Event 4: the process observes the result and continues or handles the errorInput: P1, P2, P3
Resource/state: limited
Rule: apply the Storage policy step by step
Output: show which process runs, waits, completes, or is denied
Always write the before-state and after-state for each step.Seek time, rotational delay on HDDs, transfer time, and queueing all contribute.
They have no moving head, so physical location has far less effect on access latency.
Depending on the level, it can improve speed, availability, or both, but it is not a backup.
Explore 500+ free tutorials across 20+ languages and frameworks.