SysPad › Lessons › storage lessons › S3

How S3 works

Virtually unlimited object storage. 5,500 GET/s and 3,500 PUT/s per prefix - use prefix randomisation or S3 Express One Zone for higher throughput. 11 nines durability.

S3 is where the cloud keeps its stuff: images, backups, data lakes, static sites, logs. But it’s not a hard drive, it’s object storage, a giant key-to-blob store reached over HTTP, built to hold effectively unlimited data with eleven-nines durability. Understanding "object, not file" is the key that unlocks the rest.

What you will learn

How is object storage different from a filesystem, and why does that matter?
Why are S3 "folders" an illusion, and what are they really?
How does S3 achieve such extreme durability?
How do storage classes and lifecycle rules cut cost?

Objects in buckets, not files on a disk

S3 stores objects (the bytes plus metadata) in buckets, each object addressed by a unique key. You don’t mount it like a drive; you PUT and GET objects over an HTTP API.

Objects are immutable, you replace one by overwriting its key. This model is what lets S3 scale to trillions of objects.

The folder illusion

S3’s namespace is flat. A key like img/cat.jpg has no real "img" directory, the whole string is the key, and the slash is just a character. Consoles render slashes as folders for convenience, but underneath there are only keys and prefixes.

Durability by redundancy

S3 is designed for 99.999999999% (eleven nines) durability: it automatically stores each object redundantly across multiple Availability Zones. Lose a disk, a server, even a whole AZ, your data is still safe.

⚓ The default home for backups, data lakes, media, logs, and static website assets.

Storage classes and lifecycle

Not all data is accessed equally, so S3 offers storage classes: Standard (hot), Standard-IA / One Zone-IA (infrequent access, cheaper storage, retrieval fee), Intelligent-Tiering (auto-moves based on access), and Glacier classes (archival, cheapest storage, slower/cheaper retrieval).

Lifecycle rules automate this: e.g. move objects to IA after 30 days, to Glacier after 90, delete after a year, without you touching them.

Flashcards

How is S3 different from a mounted filesystem?: It’s object storage: blobs addressed by key in a bucket, accessed over HTTP (PUT/GET), not a disk with directories you mount.
Is "reports/2026/q1.csv" a set of folders in S3?: No, it’s one flat key; the slashes are just characters. "Folders" are prefixes the console displays for convenience.
How does S3 reach eleven-nines durability?: By automatically storing each object redundantly across multiple AZs, so losing disks/servers/an AZ doesn’t lose data.
How do you cut S3 cost for rarely-accessed data?: Use a cheaper storage class (IA, Glacier) or Intelligent-Tiering, and lifecycle rules to auto-transition/expire objects.
What’s the trade with cold storage classes like Glacier?: Much cheaper storage, but higher retrieval cost and latency. Use for archival data you seldom read.

Sources

Open the SysPad canvas · Official S3 documentation