Cloud Parts

Block storage

A virtual hard drive you attach to one machine.

2 technologies on this page ↓
Live model
What you’re watchingThe VM makes many small, fast reads and writes to its attached volume. Periodic snapshots copy the volume’s state to durable storage.

How it works

Block storage gives a server a raw disk: a row of numbered blocks that knows nothing about files. The server formats it with a file system like ext4 or NTFS, which turns those blocks into named files and folders. It handles fast, small, random reads and writes, which is what databases and operating systems need. A volume lives in one availability zone, is usually attached to one VM at a time, and is backed up with point-in-time snapshots.

Think of it as

An external SSD plugged into one laptop. Quick and flexible, but only that laptop can use it at any moment.

Reach for it when

  • Boot disks for VMs
  • Self-managed databases that need low-latency random I/O
  • Any app that expects an ordinary local filesystem

Think twice when

A real example

A PostgreSQL server on a VM keeps its data on a 500 GB volume provisioned for 3,000 IOPS. A nightly snapshot is copied to another region for disaster recovery.

Technologies

Specific products, and what each does differently.

What each provider calls it

AWS
EBS
Google Cloud
Persistent Disk, Hyperdisk
Azure
Managed Disks
Open source & others
Ceph RBD, Longhorn, LVM

In code · bash

aws ec2 create-volume --availability-zone us-east-1a \
  --size 500 --volume-type gp3 --iops 3000
aws ec2 attach-volume --volume-id vol-0a1b2c3d \
  --instance-id i-0123456789abcdef0 --device /dev/sdf

# then, on the VM
sudo mkfs -t ext4 /dev/nvme1n1 && sudo mount /dev/nvme1n1 /data

Often confused with

Object storageFile storageVirtual machinesRelational databasesBackup & restoreObject, block or file storage?
Where it came from1956–2021

Block storage is as old as the disk drive. IBM’s RAMAC, revealed in 1956, was the size of two refrigerators and held 5 million characters, but unlike tape or punched cards it let a program jump straight to the record it wanted. Every disk since has worked the same way underneath: a row of fixed-size blocks that the computer reads and writes by number, with a file system on top deciding what the blocks mean. In 1981 Shugart Associates defined SASI, a disk drive interface that was adopted as the Small Computer System Interface (SCSI) and published as an ANSI standard in 1986.

Two ideas then shaped the block storage clouds sell. In 1987 David Patterson, Garth Gibson and Randy Katz at Berkeley proposed RAID: arrays of cheap PC disks, with redundancy so the array survives a failed drive, could outperform the large, expensive disks of mainframes. And the disk moved onto the network: storage area networks gave servers block devices over a dedicated network, and iSCSI (2004) carried SCSI commands over ordinary TCP/IP.

Amazon EBS (2008) turned the network disk into a self-service product: create a volume through an API, attach it to a VM, snapshot it to object storage. Every major cloud now offers the same model.

Why it’s built this way

Numbered blocks, with no idea of files
Keeping the device ignorant of files keeps it simple and lets anything sit on top: any file system, or a database that manages its own blocks.
Redundancy across many cheap disks
Single large disks were expensive and still failed. Spreading data with redundancy over many inexpensive drives made storage both faster and able to survive a dead disk.
Moving the disk onto the network
A disk on the network can outlive its server, be reattached to another one and be snapshotted centrally. The cost is a network round trip on every read and write.
Usually one machine at a time
A file system such as ext4 assumes it alone controls the disk, so two machines writing to one volume corrupt it unless they run a cluster-aware file system. Sharing files is file storage’s job.

Timeline

  1. 1956

    IBM’s RAMAC introduces random-access disk storage: 5 million characters

  2. 1986

    SCSI-1 standard published, growing out of Shugart’s SASI (1981)

  3. 1988

    Patterson, Gibson and Katz publish “A Case for Redundant Arrays of Inexpensive Disks”

  4. 2004

    iSCSI (RFC 3720) carries SCSI block commands over TCP/IP

  5. 2008

    Amazon EBS makes network block volumes self-service in the cloud

  6. 2021

    EBS io2 Block Express reaches 256,000 IOPS per volume

Sources14

Improve this page on GitHub