Block, file, object: the only three kinds of cloud storage

AWS storage looks like a wall of service names. The shortcut: there are only three kinds of storage, and every service is one of them with a brand name.

Cloud storage looks like a wall of service names the first time you meet it: EBS, EFS, FSx, S3, Glacier, and more. Here is the shortcut that makes it manageable. There are only three kinds of storage. Every service is one of them wearing a brand name.

Block: a disk for one machine

Block storage is a raw disk, the same idea as the drive inside your laptop. It attaches to one server, that server puts a filesystem on it, and it is fast. On AWS this is Amazon EBS. It lives in a single Availability Zone and belongs to one EC2 instance at a time, the way a hard drive belongs to one computer.

File: a shared drive for many machines

File storage is the shared network drive your office used to have, the one everyone mapped and dropped documents into. Many machines mount it at once and see the same folders. On AWS that is Amazon EFS for Linux, or Amazon FSx for Windows and specialised workloads. Unlike EBS it is regional, and it grows on its own as you add files.

The classic trick question is EBS versus EFS, and it is really one question: how many machines need it at once? One means EBS. A fleet means EFS.

Object: a web locker for anything

Object storage is the odd one if you grew up on files and folders, and also the most useful. You hand over a whole object (the file plus some metadata), you get back an ID or URL, and you fetch it later by that address. There are no real folders, it scales to effectively unlimited size, and you reach it over the web. This is Amazon S3, and it sits behind a huge share of what runs on AWS: backups, static websites, media, app uploads, data lakes.

• EBS is a disk for one server.
• EFS and FSx are a shared drive many servers use at once.
• S3 is a web locker you reach by URL, with room for anything.

S3 has a cost dial

One more idea worth stealing from S3: not all stored data is equal. Files you read constantly should sit somewhere instant. Files you might need once a year can sit somewhere slower and far cheaper. S3 storage classes are that dial, from Standard (on the counter) to Infrequent Access (the pantry) to Glacier (the freezer in the basement). You can even set a lifecycle rule that moves old data down the tiers on its own, so last year's logs get quietly cheaper without you lifting a finger.

Block is a disk for one. File is a shared drive for many. Object is a locker for anything. Learn the three shapes and the dozen service names stop being scary.