Chapter 18 · Hard Disk and SSD Internals

Knowledge Check

10 min read

Try to answer before you open the answers. Guessing is fine; checking your guess is how you learn.

QUICK REVIEW (5 minutes)

  1. What are the three main moving parts inside an HDD?
  2. What is the average rotational latency of a 7,200 RPM drive, and how do you work it out?
  3. What is the difference between a track, a sector and a cylinder?
  4. What is 512e?
  5. What is the smallest unit an SSD can write, and the smallest it can erase?
  6. How many bits per cell do SLC, MLC, TLC and QLC store?
  7. What does TRIM do?
  8. What does the FTL do?
  9. About how fast is SATA III in real use, and what protocol does it use?
  10. What are the two things NVMe has that AHCI lacks, in terms of queues?

ANSWER KEY — DO NOT READ UNTIL YOU'VE ANSWERED

Click to reveal answers

1. Moving parts of an HDD

The platters (spun by the spindle motor), the spindle motor itself, and the actuator arm with its read/write heads (moved by a voice coil).

2. Rotational latency at 7,200 RPM

Average = half of one rotation. One rotation = 60 ÷ 7,200 seconds ≈ 8.3 ms, so half is about 4.2 ms.

3. Track, sector, cylinder

A track is one circle on a platter surface. A sector is a small piece of a track (512 bytes or 4 KB of data). A cylinder is the same track number on every surface.

4. 512e

A drive with real 4 KB physical sectors that pretends to have 512-byte sectors so older software still works. Small writes cost extra (read, change, write back).

5. Write and erase units

Write (and read): a page (a few KB). Erase: a block (many pages).

6. Bits per cell

SLC 1, MLC 2, TLC 3, QLC 4.

7. TRIM

Windows tells the SSD which addresses no longer hold useful data, so the SSD can skip them during garbage collection.

8. FTL

The Flash Translation Layer is a table that maps the PC's logical addresses to the real physical pages in flash.

9. SATA III

About 550-600 MB/s at best (6 Gbit/s link), using the AHCI protocol.

10. Queues

NVMe supports many queues (up to about 64K) and many commands per queue (up to about 64K). AHCI has one queue of 32 commands.

UNDERSTANDING QUESTIONS

Q1. Why is a hard drive so much slower at reading 1,000 small scattered files than one large file of the same total size?

Answer Each small file needs its own seek (moving the arm) and rotational wait, so a thousand files mean about a thousand of those delays, each around 10 ms or more. A large file stored in one place needs one seek and then streams at full speed, because the head barely has to move.

Q2. Why can't an SSD simply overwrite a page in place, and what does it do instead?

Answer Flash cells can only be written when they are blank, and blanking (erasing) works only on a whole block. So the SSD writes the new data to a fresh blank page, updates its FTL map to point there, and marks the old page stale. Later, garbage collection erases blocks that are mostly stale.

Q3. Why do QLC SSDs cost less per gigabyte but wear out faster than SLC?

Answer QLC stores 4 bits per cell, so each chip holds four times as much as SLC, which makes it cheaper per GB. But it must tell 16 charge levels apart instead of 2, so reads and writes are slower, and each erase cycle does damage that makes those levels harder to tell apart, so fewer cycles are safe.

Q4. Why does an SSD get slower when it is nearly full?

Answer There are few blank pages left, so the controller has little working room. Garbage collection must run right away during your writes (foreground GC), which adds copying and erasing delays, raises write amplification, and the SLC write cache may also shrink. Free space and over-provisioning prevent this.

Q5. Why does NVMe do better than SATA even though both use flash?

Answer Two reasons. The PCIe link (usually x4) has far more bandwidth than the single 6 Gbit/s SATA link. And NVMe's protocol allows many queues with many commands, one set for each CPU core, while AHCI has one queue of 32 commands. This lets the SSD's parallel flash channels stay busy and lowers CPU overhead.

SCENARIO QUESTIONS

S1. A laptop with a hard drive starts clicking and freezes after being knocked off a desk while on. What probably happened, and what should you do first?

Answer Probably a head crash or damaged heads: the shock made the head hit the platter, scratching it. Switch the PC off and stop using it, because more spinning means more scratches. Try to recover data from a backup. If there is none, a data-recovery lab is the safe option. Do not open the drive yourself.

S2. You copy a 150 GB folder to a cheap QLC SSD. It starts at several GB/s, then drops to a few hundred MB/s. Is the drive faulty?

Answer No. The drive wrote into its fast SLC-style cache first. When the cache filled, writes went directly to the slower QLC flash. This is normal behaviour for many TLC and QLC drives. After idling, the drive empties the cache and the speed returns.

S3. You buy a new M.2 SSD that fits your laptop slot, but it is not detected. Name two possible reasons.

Answer (1) The drive is NVMe but the slot is SATA-only (or the other way round), because SATA and NVMe M.2 drives can look similar. (2) The key type or the length (such as 2280 versus 2242) does not match, or the slot is disabled or shared with another port in the BIOS. Check the motherboard or laptop manual for what the slot supports.

S4. A friend runs the old-style "defragment" on his SSD every week to "keep it fast". Is this a good idea?

Answer No. An SSD has no seek time, so scattered data costs almost nothing, and defragmenting adds extra writes (wear). What SSDs need is TRIM (retrim), which Windows does automatically through Optimize Drives. Tell him to leave the default schedule on.

S5. A NAS with RAID is rebuilding and one new drive is far slower than the others during long writes. The drive is labelled as a cheap desktop model. What might be the cause?

Answer It may be an SMR drive. SMR drives write to a small cache area first and then reorganise overlapping tracks, so during long sustained writes (like a RAID rebuild) the cache fills and the speed falls sharply. For NAS and RAID use, choose CMR drives, and check the maker's data sheet.

INTERVIEW QUESTIONS

I1. Explain how an SSD can store data without any moving parts.

Answer It uses NAND flash cells, tiny transistors with an insulated layer that traps electric charge. The amount of trapped charge represents bits and stays without power. A controller reads the charge level to get the data back, and writes by adding charge. Cells are grouped into pages (write unit) and blocks (erase unit).

I2. What is write amplification and how can you reduce it?

Answer It is the ratio of data actually written to flash divided by data the host asked to write. It exceeds 1 because garbage collection copies still-valid pages before erasing blocks. It is reduced by TRIM, more over-provisioning or free space, sequential writes, and good controller firmware.

I3. What is the difference between TBW and DWPD?

Answer TBW is the total terabytes the maker says you can write over the drive's life. DWPD is how many full-drive writes per day it can take during the warranty. They are linked: TBW is roughly DWPD times capacity times the days of warranty. Both are guarantees rather than hard limits.

I4. How do AHCI and NVMe differ in how they queue commands?

Answer AHCI has a single queue of up to 32 commands, shared by every CPU core. NVMe supports up to about 64K queues each holding up to about 64K commands, so each core can have its own queue and interrupt (MSI-X). This removes locking, reduces delay and lets the many internal flash channels stay busy.

I5. How do you check from Windows whether a drive is an SSD or HDD, and whether TRIM is on?

Answer Task Manager's Performance tab (newer builds show the type), `Get-PhysicalDisk` (MediaType and BusType), or Optimize Drives (media type column). For TRIM, run `fsutil behavior query DisableDeleteNotify`; a result of 0 means TRIM is enabled.

PRACTICAL TASK

Make a one-page "storage card" for your PC. Using the lab results, write down for each drive: the type (HDD, SATA SSD or NVMe SSD), size, interface and protocol, health status, TRIM status, and percent free. Then add two lines: the fastest your drive could possibly go (using the interface table in Section 4) and one thing you would do to look after it (for example, keep 10-20% free, or check your backup).

Why This Matters

When someone says "my PC is slow", the drive is one of the first things to check. If you can quickly say "it is an HDD at 100% active time" or "it is a SATA SSD that is 95% full", you will find the real cause quickly instead of guessing. It also helps you decide whether upgrading would really help.

CHAPTER SUMMARY

  • An HDD stores data magnetically on spinning platters. Heads on a voice-coil actuator float on an air cushion (much thinner than a hair) and read tracks divided into sectors.
  • HDD delays come from seek time plus rotational latency (about 4.2 ms on average at 7,200 RPM). Random access is slow; sequential is good. HDDs hate shocks. SMR drives can slow down in long writes.
  • An SSD has no moving parts. It stores charge in NAND flash cells (SLC to QLC bits per cell, stacked in 3D layers). Data is written in pages and erased in blocks.
  • The controller and FTL map logical addresses to physical pages, use DRAM or HMB, and a fast SLC cache that can run out.
  • Garbage collection, wear leveling, TRIM and over-provisioning keep an SSD fast and long-lived. Write amplification and TBW/DWPD describe wear; SMART reports health.
  • SATA III (about 550-600 MB/s, AHCI with 32 commands) limits SATA SSDs. NVMe over PCIe x4 (about 3.5 / 7 / 14 GB/s for 3.0 / 4.0 / 5.0) with up to 64K queues is much faster. Check M.2 type and keys in the manual.
  • Windows tools: Task Manager, Get-PhysicalDisk, Get-StorageReliabilityCounter, fsutil, Optimize Drives, Disk Management and Device Manager.