The problem with autoscaling DynamoDB is that every time you scale up they shard your table into more partitions but they never merge them back when you scale down. You end up with less throughput per partition once you've scaled back down. The only way to correct this is to completely recreate your table. Does Neptune take this into consideration?