A company is developing a solution to analyze images. The solution uses a 50 TB reference dataset and analyzes images up to 1 TB in size. The solution spreads requests across an Auto Scaling group of

Creating a new EBS volume for each instance and copying 50 TB of data would be extremely expensive due to data replication and complex to manage, failing to provide seamless *shared* data access and not being cost-effective.

While Mountpoint for S3 allows S3 access, FSx for Lustre offers a fully POSIX-compliant, high-performance shared file system experience across multiple instances, which is more robust for addressing "disk errors" related to concurrent file system access than a FUSE-based object client.

Moving the 50 TB reference data to Amazon S3 provides scalable and durable object storage, and then creating an Amazon S3-backed Multi-AZ Amazon FSx for Lustre file system allows multiple EC2 instances to concurrently access the data with high performance via a shared POSIX file system, resolving the "disk errors" from EBS sharing and ensuring seamless data reading. FSx for Lustre acts as a high-performance cache for S3, optimized for compute-intensive workloads.

Local instance storage is ephemeral and does not provide a shared, persistent dataset across an Auto Scaling group, meaning data would be lost upon instance termination and would need to be re-copied for each new instance, making it unsuitable for a shared reference dataset.