I am actually working on a big data project and I am using SSDB. The problem is, it doesn't support clustering and sharding so I might have to implement it by myself; which is not the best solution I think.

I mostly have to deal with floats or integers and the schema change really often so key/value storage is really suitable for my project but I can't store data into RAM because I generate 20GB of data per day (with 100K write/upserts per seconds).

Is there anyone of you who is using a persistent key/value storage in production environment, which can support clustering and sharding? I have tried Redis but I will run out of RAM in few days.