AWS Storage
AWS storage falls into three shapes, and almost every design question in this area is really a question about which shape the workload needs.
- Block storage presents a raw disk to a single machine. The machine puts a file system on it. Amazon EBS and EC2 instance store are the block services.
- File storage presents a shared file system that many machines mount at once over a network protocol. Amazon EFS (NFS) and the Amazon FSx family (SMB, Lustre, NFS, ONTAP) are the file services.
- Object storage presents a flat namespace of whole objects addressed by key, reachable over HTTPS from anywhere. Amazon S3 is the object service.
This section holds the reference for each of them:
- EBS volumes — volume types, performance, durability, snapshots and encryption.
- File sharing — EFS and the FSx family.
- S3 — buckets, storage classes, access control and archival.
- Instance storage — ephemeral local disk.
- Snapshots and AWS Backup — point-in-time copies and centralised backup policy.
Which service for which workload
Section titled “Which service for which workload”| Need | Service |
|---|---|
| A durable disk for one EC2 instance: boot volume, database data directory | EBS |
| Scratch space, caches or buffers that can be lost when the instance stops | Instance store |
| A shared POSIX file system for a fleet of Linux instances, across Availability Zones | EFS |
| An SMB file share for Windows applications, joined to Active Directory | FSx for Windows File Server |
| A high-performance parallel file system for HPC, analytics or model training | FSx for Lustre |
| NetApp ONTAP or OpenZFS semantics as a managed service | FSx for NetApp ONTAP, FSx for OpenZFS |
| Objects served over HTTP, data lakes, static sites, backups, archives | S3 |
| Long-term archive at the lowest price per GB | S3 Glacier storage classes |
Why EBS is not a SAN
Section titled “Why EBS is not a SAN”A traditional storage area network is built for sharing: several servers attach to the same volume through a fabric, and the array arbitrates concurrent access. EBS deliberately does not work that way. A volume is attached to one instance, in one Availability Zone, and the instance owns it outright. That constraint is what makes EBS predictable — there is no distributed lock manager to tune and no split-brain to reason about.
The exception is EBS Multi-Attach, which allows a Provisioned IOPS volume to be attached to several instances at once. It only shifts the problem: the application still needs a cluster-aware file system, because ordinary file systems corrupt data when two servers write to the same volume.
When several machines genuinely need the same data at the same time, a file service (EFS or FSx) or an object store (S3) is the right answer, not a shared block device.
Storage architecture background
Section titled “Storage architecture background”The AWS services map onto three long-standing on-premises patterns.
- DAS (direct attached storage) is a disk wired to one server. Simplest to reason about, impossible to share. EC2 instance store is DAS.
- NAS (network attached storage) is a file server on the LAN that several machines mount. Access is at file level, over NFS or SMB. EFS and FSx for Windows File Server are managed NAS.
- SAN (storage area network) is a dedicated network giving servers block-level access to storage arrays, usually over Fibre Channel or iSCSI. EBS occupies this niche in AWS, without the sharing.
The protocols that go with them:
- NFS (Network File System) originated at Sun Microsystems in 1984 and is the standard file-sharing protocol in Unix and Linux environments. EFS speaks NFS v4.1.
- SMB (Server Message Block, formerly CIFS) is the equivalent in Windows environments. FSx for Windows File Server speaks SMB and supports DFS namespaces.
- iSCSI carries SCSI commands over IP and presents a remote disk that looks local. It operates below the file-sharing protocols, at block level.
The practical difference is where the file system lives. With NFS and SMB the server owns the file system and clients send file operations. With iSCSI and EBS the client owns the file system and sends block operations — which is why a block volume cannot safely be shared without extra software.
Moving data into AWS
Section titled “Moving data into AWS”- AWS DataSync is the general-purpose online transfer service. It moves data between on-premises storage (NFS, SMB, HDFS, object stores) and S3, EFS or FSx, over the internet or AWS Direct Connect, with verification and scheduling built in. AWS names it first for new bulk-transfer work.
- AWS Storage Gateway is for hybrid access rather than one-off migration: an on-premises appliance that presents file, volume or virtual tape interfaces locally while storing the data in S3.
- The AWS Snow Family — physical devices shipped to a site and posted back — is being wound down. Since 7 November 2025 Snowball Edge devices are available only to existing customers, and AWS will discontinue support for Snowball devices in all commercial Regions on 31 December 2026. Anything written today should use DataSync, AWS Data Transfer Terminal, or a partner solution instead; for edge compute, AWS Outposts.