When many Pods or many zones need the same files, EBS is the wrong tool.

Amazon EFS

ReadWriteMany

Managed NFS, mount targets in each AZ.
Any Pod in any zone can mount it.
Dynamic provisioning through access points.
Good for shared content and CMS uploads; higher latency than EBS.

Amazon FSx

high throughput

FSx for Lustre for HPC and ML training scratch space.
Other FSx flavours for specific protocols.
Use when throughput or POSIX semantics beat EFS.

Mountpoint for S3

object as files

CSI driver that mounts a bucket for read-heavy workloads.
Not a full POSIX filesystem: limited rewrites and renames.
Ideal for large datasets and model artefacts.

Rule of thumb

One Pod, fast block	EBS gp3, RWO
Many Pods share files	EFS, RWX
Training scratch, HPC	FSx for Lustre
Large read-mostly datasets	S3 via Mountpoint

Backup and DR

Volumes are not backups. Protect data with snapshots, AWS Backup where it supports your resources, or Velero, and rehearse a restore. Deck 6 covers backup and disaster recovery in depth.

Shared and object storage: EFS, FSx and Mountpoint for S3

EBS is block storage for one node at a time. When several Pods need the same files, possibly in different zones, the choices on AWS are Amazon EFS, FSx and S3-backed mounts. EFS is a managed NFS filesystem with mount targets in each AZ, so any Pod can mount it, giving ReadWriteMany. The EFS CSI driver supports dynamic provisioning through access points, which gives each claim an isolated directory with its own POSIX identity. Latency is higher than EBS, so it suits shared content more than databases.

FSx for Lustre provides very high throughput for HPC and machine-learning training, often linked to an S3 bucket. Mountpoint for Amazon S3 exposes a bucket as a filesystem for read-heavy workloads such as loading datasets or model artefacts. It is not a complete POSIX filesystem, so check which operations your application needs.

Match the storage to the access pattern with the rule-of-thumb table. Then think about protection. A volume is not a backup: accidental deletion, corruption and regional failures need snapshots, AWS Backup coverage (where available for your resource types) or Velero, and a rehearsed restore. Deck 6 develops RPO and RTO targets and drills.

Service capabilities evolve quickly, so confirm current features, quotas and pricing before committing a design.