Consider ToroDB

[cmc333333]

I haven't used it myself, but ToroDB looks like it'd solve some of the pain points I remember from trying to model data via MongoDB. In particular, it'd avoid the field-name hashing scheme, reduce disk use, and give you a path towards a tabular data store (which seems appropriate for the types of queries Qu performs).

[jonathanwcrane]

I just saw something today that's relevant on reddit:

https://www.reddit.com/r/programming/comments/4l24hn/scaling_to_100m_mysql_is_a_better_nosql/

The top comment is pretty interesting.

The upshot of it is, MySQL probably outperforms MongoDB for the kinds of workloads Qu sees.

If we were going to build it on AWS and needed more performance than what MySQL does, AWS has a variant called Aurora that uses the same drivers (so exactly the same code) which is about 5 times as performant as MySQL.