Executive Summary

Objective: Give practitioners one reference that covers every major class of mass storage in use today: how each one physically stores bits, which interfaces and form factors carry it, where it fits in a modern architecture, and how to identify, monitor, and sanitize it from the CLI.

Target audience: Network and security engineers, sysadmins, homelabbers, and anyone who has ever had to answer “what kind of drive is actually in that box?” during a hardware refresh, an RMA, or a data disposal audit.

The short version: Mass storage falls into four physical families (magnetic, solid-state, optical, and emerging media like glass and DNA), carried over a handful of interfaces (SATA, SAS, NVMe/PCIe, USB), and deployed in three architectures (DAS, NAS, SAN) plus object and cloud storage on top. Every family trades speed, capacity, cost, and durability differently, and the right answer for any workload is almost always a mix.

The Storage Hierarchy

Before we dig into individual media, it helps to see the whole stack. The closer a tier sits to the CPU, the faster and more expensive per terabyte it is. The farther away, the cheaper, denser, and slower it gets.

 

Rule of thumb: each step down the pyramid is roughly 10x to 1000x slower, and noticeably cheaper per terabyte. That gap is why tiering exists.

Part 1: Magnetic Storage

Magnetic media stores data as the orientation of tiny magnetized regions. It is the oldest mass storage family still in production, and it still holds the bulk of the world’s data.

Hard Disk Drives (HDD)

An HDD stores data on spinning platters coated with a magnetic film. A read/write head floats nanometers above the surface on an actuator arm. Access time is dominated by mechanics: the arm has to seek to the right track, then wait for the right sector to rotate under the head.

Key components and specs:

• Platters: aluminum or glass substrates, stacked (current high-capacity drives use 10 or 11 platters).
• Spindle speed: 5,400 RPM (low power, archive), 7,200 RPM (mainstream enterprise), 10K and 15K RPM (legacy performance, now largely replaced by SSD).
• Helium fill: sealed helium drives reduce turbulence and drag, allowing more, thinner platters in the same 3.5-inch shell.
• Dual actuator (Seagate MACH.2): two independent arms on one spindle, roughly doubling sequential throughput. The drive presents itself as two logical units.

Recording technologies you need to know:

SMR comes in three flavors, and the difference matters:

• Drive-managed (DM-SMR): the drive hides the shingling behind a CMR-looking interface. The OS cannot tell, and performance tanks under sustained random writes.
• Host-aware (HA-SMR): backward compatible, but exposes zone information so the host can optimize.
• Host-managed (HM-SMR): the host must write sequentially within zones (Zoned Block Device). Used by hyperscalers with software built for it.

Where capacity stands in 2026: Seagate launched 44TB enterprise drives on its HAMR-based Mozaic 4+ platform in March 2026, qualified with two hyperscale cloud customers and ramping to volume. Western Digital is pushing ePMR with UltraSMR to 40TB in the second half of 2026, with its own HAMR drives (36TB CMR, 44TB UltraSMR) expected to ship in volume in 2027. Both vendors have published roadmaps targeting 100TB drives around the end of the decade.

Magnetic Tape (LTO)

Tape stores data on a long ribbon of magnetic film wound in a cartridge. It is sequential: to read a file near the end, the drive has to wind there first. That makes it terrible for random access and fantastic for everything else archive-related.

Why tape refuses to die:

• Cost per TB: the lowest of any mainstream medium.
• Air gap: a cartridge sitting on a shelf or in an offline library slot cannot be encrypted by ransomware.
• Longevity: LTO media is rated for decades of archival life under proper storage conditions.
• Power: an idle cartridge draws zero watts.
• WORM: write-once cartridges support compliance retention.

LTO generations at a glance:

LTO-10 gotcha: LTO-10’s new servo and track layout breaks backward read compatibility with older generations. If you have an LTO-8 or LTO-9 archive, keep a drive that can read it before you retire your old library.

LTFS (Linear Tape File System) lets a tape present itself as a mountable filesystem with a directory tree, so you can drag and drop files without proprietary backup software.

Legacy Magnetic Media

Floppy disks (8-inch, 5.25-inch, 3.5-inch), Zip and Jaz cartridges, and older tape formats (DLT, DDS/DAT, AIT) are obsolete, but you will still find them in OT/ICS environments, legacy CNC machines, and records retention closets. Treat them as data-bearing media during disposal.

Part 2: Solid-State Storage

Solid-state storage has no moving parts. Data lives as electrical charge trapped in NAND flash cells, managed by a controller that hides the messy physics behind a clean block interface.

How NAND Flash Works

Each NAND cell holds charge in a charge trap (modern 3D NAND) or floating gate (older planar NAND). The amount of charge represents one or more bits. Cells are organized into pages (the unit you write) and blocks (the unit you erase). You cannot overwrite a page in place; the drive must write somewhere new and erase old blocks later.

Cell types: bits per cell vs. endurance

P/E cycles = program/erase cycles. Figures vary widely by vendor, process node, and controller, so always check the drive’s rated TBW (terabytes written) or DWPD (drive writes per day).

The Controller and FTL: Where the Magic Happens

The SSD controller runs a Flash Translation Layer (FTL) that maps logical block addresses to physical NAND pages. It handles:

• Wear leveling: spreads writes across all cells so no block dies early.
• Garbage collection: consolidates valid pages and erases stale blocks in the background.
• Over-provisioning: hidden spare capacity that gives garbage collection room to work and replaces worn blocks.
• TRIM / UNMAP / Deallocate: the OS tells the drive which blocks are no longer in use, so garbage collection does not waste effort preserving deleted data.
• SLC caching: TLC and QLC drives often write incoming data to a fast SLC-mode region first, then fold it into dense cells later.
• Power Loss Protection (PLP): capacitors on enterprise drives flush in-flight data to NAND if power drops.

Security implication: because the FTL decides where data physically lands, overwriting a file or even the whole logical drive does not guarantee the old physical pages are gone. That is why SSD sanitization uses firmware commands, not overwrite tools. More on that below.

SSD Interfaces and Protocols

The protocol matters as much as the wire. AHCI was built for a single spinning head; NVMe was built for massively parallel flash and talks to the CPU directly over PCIe with far less overhead.

SSD Form Factors

M.2 gotcha: M.2 is a connector shape, not a protocol. An M.2 SATA drive in an NVMe-only slot will not be detected, and vice versa. Check the keying and the motherboard manual.

Where capacity stands in 2026: Micron began shipping its 245.76TB 6600 ION, a PCIe Gen5 QLC drive, in May 2026. KIOXIA, SK hynix, Sandisk, and others have 245TB-class drives announced or sampling. These are data center parts aimed at AI and cloud capacity tiers, not desktop drives.

Removable and Embedded Flash

Not all flash lives in an SSD enclosure.

Security note: firewalls, switches, and many appliances (including FortiGate models with onboard flash or SSD log disks) carry eMMC or SSD storage that holds configs, certificates, and logs. Those devices need a sanitization plan at end of life just like servers do.

Part 3: Optical Storage

Optical media stores data as microscopic pits and lands (pressed discs) or as changes in a dye or phase-change layer (recordable discs), read by a laser.

Where optical stands today: consumer optical is in decline, and major manufacturers have been winding down recordable media production. Optical still has a niche in compliance archives (WORM by physics), software distribution in air-gapped environments, and evidence handling, but tape and object storage have taken most of its archive role.

Part 4: Hybrid and Persistent-Memory Storage

These technologies sit between categories or between tiers.

• SSHD (Solid-State Hybrid Drive): an HDD with a small NAND cache that holds frequently read blocks. Cheap speedup for consumer systems, now largely displaced by plain SSDs.
• Fusion Drive (Apple): macOS paired a separate SSD and HDD into one logical volume with automatic tiering. Apple has since dropped it in favor of all-flash.
• OptiNAND (Western Digital): embeds flash inside high-capacity HDDs for metadata, which frees platter space and improves capacity and resilience.
• Intel Optane (3D XPoint): a non-NAND persistent memory with very low latency and enormous endurance, sold as SSDs and as DIMMs. Intel announced the wind-down of the Optane business in 2022, but you will still find it in installed servers.
• NVDIMM: DRAM backed by flash and a power source so memory contents survive power loss.
• CXL memory expansion: Compute Express Link lets servers attach extra memory over PCIe-based links, often in E3.S form factors. It is memory, not mass storage, but it blurs the line between the top two tiers of the pyramid.

Part 5: Storage Architectures

The medium is only half the story. How storage is attached and shared defines how you secure, back up, and scale it.

DAS (Direct-Attached Storage)

Drives connected directly to one host: internal bays, a USB or Thunderbolt enclosure, or a SAS JBOD. Simple and fast, but only that host can use it natively.

NAS (Network-Attached Storage)

A file server sharing files over the network using file protocols:

• SMB/CIFS: Windows-native, also used by macOS.
• NFS (v3, v4.x): Unix/Linux-native.
• AFP: Apple’s legacy protocol, deprecated.

Mount an NFSv4 export from a NAS on Linux:

sudo mkdir -p /mnt/archive
sudo mount -t nfs4 10.0.20.10:/export/archive /mnt/archive
df -h /mnt/archive

SAN (Storage Area Network)

A dedicated network that presents blocks (raw LUNs or namespaces) to hosts, which then format them with their own filesystem.

Discover and log in to an iSCSI target on Linux:

sudo iscsiadm -m discovery -t sendtargets -p 10.0.20.20
sudo iscsiadm -m node --login
lsblk

Connect to an NVMe/TCP subsystem on Linux:

sudo modprobe nvme-tcp
sudo nvme discover -t tcp -a 10.0.20.30 -s 8009
sudo nvme connect -t tcp -n <SUBSYSTEM_NQN> -a 10.0.20.30 -s 4420
sudo nvme list

Security note: iSCSI and NVMe/TCP are just TCP traffic. Put them on an isolated storage VLAN, enforce CHAP (iSCSI) or DH-HMAC-CHAP (NVMe-oF), and never let storage traffic share a flat network with user endpoints. A firewall policy between your storage segment and everything else is cheap insurance.

Object Storage

Object storage stores data as objects (data plus metadata plus a unique key) in flat buckets, accessed over HTTP APIs, with the Amazon S3 API as the de facto standard. It scales to exabytes, uses erasure coding instead of RAID, and supports features like versioning and Object Lock (immutable, WORM-style retention) that are core to modern ransomware-resilient backup designs.

Examples: AWS S3, Azure Blob, Google Cloud Storage, plus self-hosted options like Ceph RGW and MinIO-compatible platforms.

Cloud Storage Tiers

Public cloud wraps all of the above into service tiers: block (EBS, Azure Managed Disks), file (EFS, Azure Files), object (S3), and archive classes (S3 Glacier Deep Archive, Azure Archive) that are often backed by tape or similar cold media under the hood. The trade-off in archive tiers is retrieval time and retrieval fees.

Software-Defined Storage

Platforms like ZFS, Ceph, and Windows Storage Spaces pool commodity drives and handle redundancy, snapshots, and checksumming in software rather than on a hardware RAID controller.

Part 6: RAID and Redundancy Quick Reference

RAID is not a storage type, but it is how most mass storage is deployed, and the choice of medium changes the right RAID level.

Large-drive warning: rebuilding a failed 30TB to 44TB HDD can take days. During that window a second failure or an unrecoverable read error can lose the array. With large HDDs, use dual parity (RAID 6, RAIDZ2) or better, and keep tested backups.

RAID is not backup. It protects availability, not against deletion, corruption, or ransomware. Follow 3-2-1 (or 3-2-1-1-0: one copy offline or immutable, zero errors on restore tests).

Part 7: Emerging Storage Media

These are worth tracking even though none are in your rack yet.

• Glass storage: femtosecond lasers write data as 3D structures inside fused silica (for example, Microsoft’s Project Silica research). Designed for archive lifetimes measured in centuries with no need for climate control.
• DNA storage: encodes bits in synthetic DNA sequences. Enormous theoretical density and longevity, but write and read speeds and costs are still far from practical.
• Holographic storage: stores data throughout the volume of a medium rather than on a surface. Repeatedly promising, repeatedly not commercial.
• Ceramic and other novel archive media: several startups are pursuing ultra-durable write-once archive media aimed at the cold tier currently owned by tape.

The Big Comparison Table

Identifying and Monitoring Storage from the CLI

Linux

List block devices and see whether each one is rotational (ROTA=1 is an HDD, 0 is flash) and how it is attached:

lsblk -d -o NAME,ROTA,TRAN,SIZE,MODEL

Check SMART health on SATA/SAS and NVMe:

sudo smartctl -a /dev/sda
sudo smartctl -a /dev/nvme0

NVMe-specific inventory and wear (watch percentage_used and media_errors):

sudo nvme list
sudo nvme smart-log /dev/nvme0

Check whether the kernel sees a drive as zoned (host-aware or host-managed SMR):

cat /sys/block/sda/queue/zoned

Gotcha: drive-managed SMR reports none here, the same as CMR. The only reliable way to know is the model number and the vendor datasheet.

Tape drives and changers:

lsscsi -g
sudo mt -f /dev/nst0 status

macOS

diskutil list
diskutil info disk0 | grep -E "Device / Media Name|Protocol|Solid State"
system_profiler SPNVMeDataType

Windows (PowerShell)

Get-PhysicalDisk |
  Select-Object FriendlyName, MediaType, BusType, HealthStatus,
    @{n='SizeGB';e={[math]::Round($_.Size/1GB)}}

Get-PhysicalDisk | Get-StorageReliabilityCounter |
  Select-Object DeviceId, Temperature, Wear, ReadErrorsTotal

FortiGate

Many FortiGate models have an internal SSD or flash disk used for logging and WAN optimization caching. Check what is installed and its status:

execute disk list
diagnose hardware deviceinfo disk

Sanitization: Getting Data Off Each Media Type

Every storage type needs a different end-of-life approach. NIST published SP 800-88 Revision 2 in September 2025, replacing the 2014 Rev. 1. It keeps the three sanitization methods (Clear, Purge, Destroy), shifts focus to building an organizational sanitization program, and points to standards like IEEE 2883 for media-specific techniques instead of maintaining its own per-device tables. It also explicitly discourages the old multi-pass overwrite approach for HDDs.

Degaussing does nothing to flash. SSDs, USB drives, and SD cards store charge, not magnetism.

Check NVMe sanitize support, then run a crypto erase sanitize:

sudo nvme id-ctrl /dev/nvme0 -H | grep -i -A4 "sanicap"
# WARNING: destroys all data on the controller's namespaces
sudo nvme sanitize /dev/nvme0 --sanact=4
sudo nvme sanitize-log /dev/nvme0

--sanact values: 2 = block erase, 3 = overwrite, 4 = crypto erase.

SATA drive sanitize (check status first; the drive must not be “frozen”):

sudo hdparm -I /dev/sdX | grep -i -A8 "Security:"
# WARNING: destroys all data
sudo hdparm --yes-i-know-what-i-am-doing --sanitize-crypto-scramble /dev/sdX
sudo hdparm --sanitize-status /dev/sdX

Verify, then document. Rev. 2 puts weight on validating that sanitization worked and recording it (media type, serial, method, date, operator) in a certificate of sanitization.

Troubleshooting and Gotchas

1. SMR drives in RAID or NAS arrays. Symptom: rebuilds that take days, arrays that drop drives as “failed” under heavy writes. Cause: drive-managed SMR drives stall when their CMR cache region fills. Fix: check the model against the vendor’s CMR/SMR list before buying. Use CMR drives for any parity RAID or ZFS pool.

2. QLC write cliff. Symptom: a large copy starts at several GB/s and then drops to a few hundred MB/s or less. Cause: the SLC cache filled, and the drive is now writing directly to QLC. Diagnose: run a sustained write longer than the cache size and watch throughput:

fio --name=seqwrite --filename=/mnt/test/fio.bin --rw=write \
    --bs=1M --size=200G --direct=1 --ioengine=libaio --iodepth=16

Fix: size QLC for read-heavy data; put write-heavy workloads on TLC or SLC-cached tiers.

3. Consumer SSDs in servers. Symptom: corruption after power events, unexpectedly fast wear-out. Cause: no power loss protection and low DWPD ratings. Fix: use enterprise SSDs with PLP for databases, hypervisors, and ZFS SLOG devices. Monitor wear with nvme smart-log.

4. Counterfeit capacity on flash media. Symptom: a “2 TB” USB stick or microSD card that silently corrupts data after a few dozen GB. Cause: firmware lies about capacity. Diagnose:

sudo f3probe --destructive --time-ops /dev/sdX

Fix: buy from authorized channels and test new media before trusting it.

5. LTO generation mismatches. Symptom: an older tape will not load or read in a new drive. Cause: each LTO generation reads only a limited range of older generations, and LTO-10 breaks the pattern entirely. Fix: keep a compatible drive (and a documented restore test) for every generation still in your retention schedule.

How to Choose: Quick Decision Guide

• Boot drive, VMs, databases: NVMe TLC SSD, enterprise class with PLP for servers.
• Large read-heavy datasets (AI, media libraries, analytics): NVMe QLC SSD if performance matters, CMR HDD if cost matters.
• NAS and backup targets: CMR HDDs in RAID 6 / RAIDZ2, plus an SSD cache if needed.
• Immutable backup copy: object storage with Object Lock, or LTO tape stored offline.
• Long-term cold archive: LTO tape, or a cloud archive tier.
• Transport and field work: USB flash or SD, encrypted, never the only copy.
• Compliance WORM on a small scale: LTO WORM cartridges or archival optical.

Key Takeaways

• Mass storage comes in four physical families: magnetic (HDD, tape), solid-state (NAND flash in every form factor), optical, and emerging archive media.
• Interface and protocol are separate from medium. M.2 is a shape, NVMe is a protocol, PCIe is the bus, and TLC is the cell type. Know all four when you spec a drive.
• HDDs are not going away. HAMR has pushed them to 44TB, and vendors are targeting 100TB. They remain the cheapest online capacity.
• SSDs have taken over performance and are now chasing capacity, with 245TB-class QLC drives shipping in 2026.
• Tape is the air gap. LTO-10 holds 30TB or 40TB native per cartridge, but plan around its broken backward compatibility.
• Sanitize by media type. Overwrites do not reliably clean SSDs, degaussing does nothing to flash, and NIST SP 800-88 Rev. 2 is now the reference.

Further Reading

• NIST SP 800-88 Rev. 2, Guidelines for Media Sanitization
• LTO Program (lto.org)
• Micron 6600 ION 245TB review (StorageReview)
• Western Digital HAMR roadmap (Tom’s Hardware)
• Seagate Mozaic 4+ 44TB launch