Typically no. You're actually limited on the number of buckets you can create whereas number of objects, and thus prefixes, effectively not. The situation where you want different buckets is where you want to specify different bucket policies; e.g., for data lifecycle (+/- versioning, automatic archive to glacier), security, and environment (dev, test, prod). The design of prefixes/key names/directories should then be guided by your access patterns with similar sorts of considerations you have for organizing data in HDFS. Listings over prefixes/recursive listings can be slow, so thinking along those terms, if you're going to do listings you'll want enough hierarchy or structure to your key names that those result sets don't get huge. If you're only ever going to do access to specific keys, this is less of an issue.
... View more
