1. They probably don't need sharding, since the dataset is small enough to just replicate it in mirrors. 2. 1ms is achievable with SSDs, but 200K q/minute seem

1. They probably don't need sharding, since the dataset is small enough to just replicate it in mirrors. 2. 1ms is achievable with SSDs, but 200K q/minute seems slow my gut feeling tells me. This post is more like "ha we don't need NoSQL for this special use case" - Once you need scaling and some sort of atomics, you quickly have to use HBase for row-level atomicity and scaling. Redis is probably better suited for the posted usecase anyway.