4. Storage and file systems

A disk is just a long list of numbered blocks. The operating system turns it into the folders, files and drives that you use every day. In this page you will see how a disk is organised, how it is divided into partitions, which file systems you will meet (FAT32, NTFS and ext4) and why journaling can save your data after a power cut. At the end you will put it together in an installation plan.

How to use this page

1. Read sections 1 to 6 (about 70 minutes).
2. Practise in a virtual machine: Activity 6 uses GParted with a virtual disk.
3. Section 7 is the base for the installation plan of the same activity.

1. Disks

Two kinds of disk are common:

	HDD (hard disk drive)	SSD (solid-state drive)
Technology	Magnetic platters that spin and a moving head	Flash memory chips, no moving parts
Speed	Slow to find data (the head must move)	Very fast, also for random access
Weak point	Mechanical wear, shocks	Limited writes per cell (the controller spreads the wear)
Fragmentation	Hurts performance	Does not matter

The physical structure of an HDD is made of platters with two faces. Each face has circular tracks, and each track is divided into sectors. The same track on all the platters forms a cylinder. A sector is the smallest unit the disk can read or write: traditionally 512 bytes, and 4096 bytes in newer disks.

The logical structure is what the OS builds on top. The OS does not work with single sectors but with clusters (Windows) or blocks (Linux): groups of contiguous sectors. The cluster is the smallest unit assigned to a file.

This has a consequence: a file always uses a whole number of clusters, and the free space in its last cluster is wasted (internal fragmentation again). Example with clusters of 4096 bytes:

File size (bytes)	Clusters	Space used	Wasted
1,200	1	4,096	2,896
5,000	2	8,192	3,192
9,000	3	12,288	3,288
40,000	10	40,960	960
Total 55,200	16	65,536	10,336

With clusters of 32 KiB (32,768 bytes) the same four files need 5 clusters (163,840 bytes) and waste 108,640 bytes. Big clusters waste space with small files; small clusters need more clusters per file and a bigger allocation table. The cluster size is chosen when you format the partition, and the default is usually good.

2. Partitions

A partition is a part of a disk that the OS treats as if it were a separate disk. Why divide a disk?

• To install several operating systems on the same disk.
• To separate the system from your data: you can reinstall the system without touching your files.
• To use different file systems.
• To protect the system: a full data partition does not stop the OS.

The list of partitions is stored in a partition table at the beginning of the disk. There are two formats.

MBR (Master Boot Record)

It appeared in the 1980s and works with the classic BIOS firmware. The first sector of the disk (512 bytes) contains a small boot program, a table with four entries and a signature. Its limits:

• A maximum of four primary partitions. To have more, one primary partition becomes an extended partition, which is a container for any number of logical partitions. The extended partition does not store data itself.
• Partition addresses have 32 bits, so a disk is limited to 2 TiB. A bigger disk can only use the first 2 TiB.
• There is only one copy of the table. If the first sector is damaged, the partitions are lost.

GPT (GUID Partition Table)

It is the modern format, used with UEFI firmware.

• The disk has a protective MBR in the first sector, so old tools see that the disk is not empty.
• A GPT header and the partition entries (128 by default in Windows) come next; a backup copy is stored at the end of the disk.
• All partitions are the same kind (there is no extended or logical partition) and each has a unique identifier (GUID).
• It supports disks of many terabytes.
• Windows boots from a GPT disk only when the computer uses UEFI. Windows 11 requires it.

	MBR	GPT
Firmware	BIOS (legacy)	UEFI
Partitions	4 primary, or 3 primary + 1 extended with logical ones	Up to 128 in Windows, all the same type
Maximum disk	2 TiB	Practically unlimited
Table copies	1	2 (start and end of the disk)

A UEFI system needs a small EFI System Partition (ESP, formatted FAT32, usually 100 MB to 1 GB) where the boot loaders live.

Names of partitions

• Windows gives a drive letter to each volume: C:, D:…
• Linux has a single tree. Each disk appears as a device (/dev/sda, /dev/nvme0n1) and its partitions are numbered (/dev/sda1, /dev/nvme0n1p1). To use a partition you mount it in a folder, for example /home or /media/usb.

LVM

In Linux you can also use LVM (Logical Volume Manager). You join one or more disks or partitions (physical volumes) in a volume group and then cut logical volumes from it. The advantage is that you can extend a logical volume later, even by adding a new disk, without moving data.

3. File systems

A partition is empty until you format it with a file system: the set of rules and data structures that the OS uses to store and find files. A file system:

• organises files in folders (a tree that starts at the root);
• keeps track of which clusters belong to each file and which are free;
• stores the attributes (metadata) of every file: name, size, dates (created, modified, accessed), owner and permissions, and flags such as read-only or hidden;
• (in some systems) protects data with permissions, compression, encryption or a journal.

Different file systems keep their location information in different ways:

• FAT32 has a File Allocation Table. It is a table with one entry for each cluster. A file is a chain: the directory entry gives the first cluster, and the table says which cluster comes next. For example, a file in clusters 5, 6 and 9 has the chain 5 → 6 → 9 → end. There are two copies of the table.
• NTFS has the Master File Table (MFT), with at least one record for every file or folder. Very small files can be stored inside their own record.
• ext4 uses inodes: one small structure per file with its attributes and the location of its data (in extents, ranges of contiguous blocks). The name of the file is stored in the directory, not in the inode.

Comparison

	FAT32	exFAT	NTFS	ext4
Main system	Everything (old standard)	Windows, macOS, Linux	Windows	Linux
Maximum file	4 GiB − 1 byte	Very large	Limited by the volume	16 TiB
Maximum volume	2 TiB (Windows formats up to 32 GiB with its own tool)	Very large	Very large	1 EiB
Permissions	No	No	Yes (ACL)	Yes (Unix and ACL)
Journal	No	No	Yes (metadata)	Yes
Names	Not case-sensitive	Not case-sensitive	Not case-sensitive in Windows	Case-sensitive
Typical use	Small USB drives, SD cards, EFI partition	Big USB drives and SD cards for several systems	Windows system and data disks	Linux system and data disks

How to choose, with a few examples:

• A USB drive that you use in Windows, macOS and Linux and that has files bigger than 4 GiB: exFAT.
• A USB for a camera, a TV or an old device: FAT32.
• The Windows system partition, or a Windows data disk with permissions: NTFS.
• The Linux system partition: ext4.

Compatibility between systems is not complete: Windows cannot read ext4 without extra software, and macOS reads NTFS but does not write it by default. Linux can read and write NTFS, exFAT and FAT32.

4. Journaling

Saving a file needs several writes: the data blocks, the table with the free space and the directory entry. If the power goes off in the middle, the file system can be left inconsistent: a directory entry that points to nothing, or blocks marked as free and used at the same time.

Without a journal (FAT32, exFAT, old ext2), the system must check the whole disk after a bad shutdown (fsck in Linux, chkdsk in Windows). On a big disk this can take a long time, and it does not always repair everything.

A journaling (or transactional) file system keeps a journal: a small log area. Before changing the real structures, it writes in the journal what it is going to do. After that, it makes the changes and marks the entry as finished.

1. Write the planned change in the journal.
2. Commit: mark the entry as complete.
3. Apply the change to the real structures.
4. Free the entry in the journal.

After a power cut, at the next boot the system reads the journal:

• Entries that were committed but not applied are replayed (repeated).
• Entries that were not committed are discarded, as if they had never started.

The disk is consistent again in seconds, without checking everything. The price is some extra writes.

Journaling does not protect your files from everything

NTFS and ext4 (in its default ordered mode) journal mainly metadata (the structures of the file system). That keeps the file system healthy, but the last data you were writing can still be lost or incomplete. Journaling also does not replace a backup.

Consequence for choosing a file system: for the system disk and for data that matters, choose a journaling file system (NTFS or ext4). For removable drives in FAT32 or exFAT, always use Safely remove hardware or umount before pulling the cable.

5. Useful commands and tools

Task	Linux	Windows
See disks and partitions	lsblk, lsblk -f (with file systems), sudo fdisk -l	Disk Management (diskmgmt.msc), diskpart
Free and used space	df -h (by file system), du -sh folder (one folder)	File Explorer → This PC, Settings → Storage
Graphic partition editor	GParted	Disk Management
Create a file system	sudo mkfs.ext4 /dev/sdb1	Format in Explorer, or format
Check and repair	sudo fsck /dev/sdb1 (unmounted)	chkdsk D: /f

Partitioning and formatting destroy data

Do these exercises only in a virtual machine with a virtual disk, or on an empty USB drive that you do not need. Before you apply a change, read the name of the device twice: formatting /dev/sda instead of /dev/sdb can erase your system. Nothing is written until you press Apply in GParted.

6. Optimising storage

• Keep free space. A full disk is slow and risky. Aim for at least 10–15 % free on the system partition.
• Clean up. Use Storage Sense or Disk Cleanup in Windows, and in Linux du -sh * to find the big folders and sudo apt clean to delete downloaded packages (Debian and Ubuntu).
• HDD: defragment. When a file is stored in many pieces, the head needs more movements. Windows does it automatically (Defragment and Optimize Drives). ext4 fragments very little and rarely needs it.
• SSD: never defragment; use TRIM. The TRIM command tells the SSD which blocks are free so that it can prepare them. Windows does it by itself. In Linux, sudo fstrim -av runs it once, and most distributions run it every week.
• Choose the right cluster size and file system for the use: many small files or a few huge ones.
• Align partitions. Modern tools start partitions at 1 MiB, which is right for SSDs and 4 KiB sectors.
• Watch the health. If a disk starts to fail (strange noises, many read errors), back up first. In Linux, smartctl -a /dev/sda (package smartmontools) shows the SMART report.

7. The installation plan

Before you install an operating system, write a plan. It avoids errors and it is evidence of your work. These are the points to cover:

Point	Questions to answer
Purpose and users	What will the computer be used for? Who uses it?
Hardware	Are the CPU, RAM and disk enough? Are there drivers? BIOS or UEFI?
Operating system	Which one, which edition and version, 32 or 64 bits? Which licence?
Disk	Which partition table? Which partitions, sizes and file systems? Is there swap? Is there dual boot?
Network and name	Computer name, domain or workgroup, IP address (automatic or fixed)
Users and security	Administrator account, other accounts, passwords, firewall, updates
Software	Which applications and drivers will you install after the system?
Backup	What will you copy before the installation? How will you go back if it fails?
Checks	How will you test that everything works?

Example. A virtual machine for Ubuntu Desktop, with 4 GiB of RAM and a 40 GiB virtual disk (UEFI):

Partition	Size	File system	Mount point
EFI System Partition	512 MiB	FAT32	/boot/efi
Root	25 GiB	ext4	/
Swap	2 GiB	swap	(none)
Data	about 12.5 GiB	ext4	/home

Ask the manufacturer for the current minimum requirements. As a reference, Windows 11 requires a 64-bit CPU, 4 GB of RAM, 64 GB of storage, UEFI with Secure Boot and a TPM 2.0 chip, and Ubuntu Desktop recommends 4 GB of RAM and 25 GB of disk.

Comprueba lo aprendido

1. A disk uses MBR. You already have three primary partitions and need three more. How can you do it?

   Answer

   MBR allows only four primary partitions. Convert (or create) the fourth one as an extended partition and create the three new partitions as logical partitions inside it.

2. A file of 10,000 bytes is saved on a file system with 4096-byte clusters. How many clusters does it use and how much space is wasted?

   Answer

   10,000 / 4096 = 2.44, so it needs 3 clusters (12,288 bytes). The waste is 12,288 − 10,000 = 2,288 bytes.

3. You want to copy a 6 GiB video to a USB drive that you will also use on a Mac. Which file system do you choose and why not FAT32?

   Answer

   exFAT. FAT32 cannot store a file larger than 4 GiB, and exFAT works in Windows, macOS and Linux. NTFS is not written by macOS by default and ext4 is not read by Windows.

4. Explain in two sentences why a journaling file system recovers faster after a power cut.

   Answer

   Before changing the disk structures, it writes the planned change in a journal. After a failure, it only reads the journal and repeats the finished entries or discards the unfinished ones, instead of checking the whole disk.

5. Why should you not defragment an SSD, and what do you use instead?

   Answer

   An SSD has no moving head, so fragmentation does not slow it, and extra writes wear it. You use TRIM, which tells the SSD which blocks are free.

Glossary

English	Castellano
disk / hard disk (HDD) / solid-state drive (SSD)	disco / disco duro / unidad de estado sólido
platter, track, sector, cylinder	plato, pista, sector, cilindro
cluster / block	clúster / bloque
partition	partición
partition table	tabla de particiones
primary / extended / logical partition	partición primaria / extendida / lógica
MBR / GPT	MBR / GPT
boot loader	gestor de arranque
mount point	punto de montaje
format	formatear
file system	sistema de archivos
file attributes	atributos del archivo
journaling	registro por diario (journaling)
journal	diario
inode	inodo
defragmentation	desfragmentación
installation plan	plan de instalación

Credits

This page adapts the SOM course notes by Moisés Mas Rocamora, Daniel Álvarez Santamaría and Jana Taboada León, licensed under CC BY-NC-SA 4.0. This page is shared under the same licence.