> Filesystem software, especially databases, can't be ported to Amazon S3 Except they can be. You don't need to overwrite the whole DB file on every INSERT/UP

> Filesystem software, especially databases, can't be ported to Amazon S3 Except they can be. You don't need to overwrite the whole DB file on every INSERT/UPDATE/DELETE; those can be (and often are) stored in memory and periodically checkpointed. You might lose some writes if the process goes down between checkpoints, but for a lot of applications that's entirely acceptable. Indeed, for SQLite in particular there are tools like Litestream that support replication to and restoration from S3. Alternately, you could split the DB across multiple files, and then an INSERT/UPDATE/DELETE would only need to overwrite the files actually affected. This is already how server-based RDBMSs usually work.