EFS

An EBS volume attaches to one instance in one AZ. But plenty of workloads — a fleet of web servers sharing user uploads, a CMS, CI workers, container tasks — need many machines reading and writing the same files at the same time. Amazon EFS is a fully managed NFS v4 filesystem: you mount it from thousands of instances (or Lambda, or containers) concurrently, and it grows and shrinks automatically as you add or delete files. No capacity planning, no resizing.

• EFS is a Regional service. In Standard mode, data is stored redundantly across multiple AZs, so it survives a full AZ outage.
• You reach it through mount targets — one elastic network interface per AZ, each with its own IP, living in a subnet you choose. Instances mount the target in their own AZ to avoid cross-AZ data charges.
• One Zone storage classes keep data in a single AZ for ~47% lower cost, trading away the multi-AZ durability.
• Access is Linux-only (NFS). Windows workloads use FSx, not EFS.

graph TD
subgraph REGION["AWS Region"]
  EFS["EFS File System (elastic, multi-AZ)"]
  subgraph AZA["Availability Zone A"]
    MTA["Mount Target (ENI)"]
    APPA["App Instances"]
    APPA -->|"NFS v4"| MTA
  end
  subgraph AZB["Availability Zone B"]
    MTB["Mount Target (ENI)"]
    APPB["App Instances"]
    APPB -->|"NFS v4"| MTB
  end
  MTA --> EFS
  MTB --> EFS
end
classDef fs fill:#eaf7ec,stroke:#2e8b3d,stroke-width:2px,color:#1a3d20
classDef app fill:#e8f0fe,stroke:#3b82f6,stroke-width:2px,color:#173a70
class EFS fs
class APPA,APPB app

• Standard and One Zone — the durability tier (multi-AZ vs single-AZ).
• Infrequent Access (IA) and Archive — cheaper per-GB tiers with a per-access retrieval fee. Lifecycle Management moves files between tiers automatically based on last-access time. EFS Intelligent-Tiering does this for you with no retrieval charge.
• Throughput: Elastic (default, pay per use, scales to demand), Bursting (scales with size), Provisioned (fixed high throughput regardless of size).
• Performance: General Purpose (lowest latency, default) vs Max I/O (higher aggregate throughput, higher latency — legacy).

You pay per GB stored with no pre-provisioning — but EFS Standard is roughly 3× the price of gp3 EBS and ~7× the price of S3 Standard per GB. That premium buys the shared POSIX filesystem. Lifecycle policies into IA/Archive cut idle-data cost dramatically. Provisioned Throughput adds a separate MB/s charge. Cross-AZ traffic to a mount target in another AZ is billed as data transfer.

• Decision tree: one instance, block device, database → EBS. Many Linux instances, shared POSIX files → EFS. Windows / SMB → FSx for Windows. Object storage over HTTP → S3.
• EFS scales to petabytes and thousands of concurrent NFS clients automatically — “shared file storage that scales with no capacity planning” is the classic EFS trigger.
• Enable lifecycle management to IA for “reduce cost of a file share where most files are rarely touched.”
• Secure it with security groups on the mount targets, IAM authorization for NFS, and encryption at rest (KMS) + in transit (TLS).
• EFS Access Points enforce a specific POSIX user/group and a root directory per application — the standard way to share one filesystem safely between multiple apps or containers.