Currently we're using cassandra with cassandra-driver, doing standard queries and accessing "raw" results in our DAL and BL layers. Our application should support millions of users, each request in average dose 3 reads from the DB, we have about 30 models.
We are thinking about using some kind of ORM library, map the models our own and handle each raw in a representative object, or simple functions to handle the get, set and creation for each model(raw data). Our use cases are manly for encapsulating the handling of "raw" data structure, avoiding code duplications, preventing schema structure mistakes and enforcing validations. We manly wonder about the impact on performance, future flexibility and lack of ORM lib/framworks popularity.
It will be helpful to get best practices (and references) in this subject, as we don't really know what's the impact using ORM layer over cassandra.