JoeyAndres / rl

Reinforcement Learning library

2 stars 0 forks source link

Investigate backend storage. #4

Closed JoeyAndres closed 7 years ago

JoeyAndres commented 7 years ago

Consider various NoSQL db for our problem:

https://docs.mongodb.com/manual/sharding/
Apache cassandra: http://cassandra.apache.org/ (REALLY REALLY FAST and easy to split in Many nodes).
- Is eventual consistency enough for all rl problem?
- If just create an API for our end and make it seamless to connect to various db (or have a default, probaly mongo).

Edit:

Afte looking into cassandra, UDT have to be declared outside of c++, the prospect of just having to store abstract json obj in mongo is too attractive.
Plus we can make mongo eventually consistent even with sharding.
Conclusion: Stick with mongo.

JoeyAndres commented 7 years ago

7 #8