I suggest you to read a whitepaper first. The “sharding” already in place for sure 
There is a whole section “2.5 Durability, device failure, and churn” describes online requirements and math behind it, especially in the “3.4 Redundancy”, “3.8 Data repair” and so on.
Also, you can read the blog posts:
- Why (Proof-of-) Replication is Bad for Decentralized Storage, Part 2: Churn and Burn
- Reputation Matters When it Comes to Storage Nodes
- Sharing Storage Space for Fun and Profit
- Early feedback from V3 storage node operators: You’ve got questions, we’ve got answers!
You can read the design draft here: Design draft: New way to measure SN uptimes