Design High-Performing Architectures

Object storage (Amazon S3) for unstructured blobs accessed via API; file storage (Amazon EFS / FSx) for shared hierarchical filesystems mounted over NFS/SMB; block storage (Amazon EBS) for raw volumes attached to a single instance like a disk.

Object storage suits large volumes of unstructured data, web content, backups, data lakes, and any workload needing massive scale, durability, and HTTP/API access. Block storage suits low-latency, transactional reads/writes from a single instance (databases, boot volumes).

Each S3 prefix sustains at least 3,500 PUT/COPY/POST/DELETE and 5,500 GET/HEAD requests per second. To scale higher, spread objects across many prefixes so requests parallelize across them.

Multipart upload splits a large object into parts that upload in parallel, increasing throughput, allowing retry of only failed parts, and enabling resumable uploads. AWS recommends it for objects over 100 MB and requires it above 5 GB.

A byte-range fetch downloads only a specified range of bytes from an object using the Range header. It lets you parallelize downloads of large objects across ranges and retrieve just the part you need (e.g., a file header), reducing latency.

S3 Transfer Acceleration routes uploads/downloads through the nearest CloudFront edge location, then over the optimized AWS backbone to the bucket. It speeds long-distance transfers of large files over the public internet; it offers no benefit for nearby clients.