How to do horizontal scaling to a RDBMS or NoSQL database in practical

Question

Im doing a research on using RDBMS and NoSQL for enterprise applications. While researching I came across that ability to horizontal scaling is a major plus point in selecting a proper database. My question is how do we actually do this in a practical scenario.

Eg: Lets say I have a customer database with 100 tables and each contains few millions of records. How should I initiate a horizontal scaling for this kind of situation.

This is what I thought

1. Replicate the DB but then how to manage the data redundancy and how we program to query data
2. Keeping indexes about the data in the each server

Is there anything more ?

Appreciate your comments.