S3 is advertised as "infinitely scalable," but in practice performance can differ by tens of times depending on how you use it. Creating a hot partition with bad prefix design, having to start over from scratch on a network error while uploading a large file with a single PUT, or having users worldwide access a single-region S3 over the internet — in all of these situations, S3 becomes the bottleneck. This day is about understanding the mathematical limits of S3 performance and digging into the design patterns that make the most of those limits.
S3's throughput is applied independently per prefix