Course contentsShow
← System Design
Lesson 360 of 1,91910. NoSQL: Key-Value StoresFree lesson

What Makes a Key-Value Store Distributed

Understanding data partitioning across multiple nodes and how distributed KV stores differ from single-node stores.

What Makes a Key-Value Store Distributed

You'll learn: How key-value stores scale by splitting data across multiple servers and what changes when we go from one node to many.

From Single Node to Distributed

A single-node key-value store (like a standalone Redis instance) keeps all data on one machine. It's fast and simple, but limited by that machine's memory, CPU, and network capacity. When data or traffic outgrows one server, we need a distributed key-value store.

A distributed KV store spreads data across multiple nodes (servers). Instead of storing all keys on one machine, the system partitions (shards) the keyspace. For example, keys starting with A-M might live on Server 1, while N-Z live on Server 2.

The Core Difference: Data Partitioning

Data partitioning is the defining feature. The system uses a partitioning strategy (like consistent hashing or range partitioning) to determine which node stores each key. When a client requests GET user:12345, the system calculates which node owns that key and routes the request there.

Think of it like a library system: A single-node store is one building with all books. A distributed store is multiple branches—when you request a book, the librarian routes you to the correct branch based on the call number.

Additional Distributed Characteristics

Beyond partitioning, distributed KV stores typically add:

  • Replication: Each key is copied to multiple nodes for fault tolerance (like Redis Cluster or Redis Replication you've learned)
  • Coordination: Nodes communicate to handle failures and rebalancing
  • Routing logic: Clients or proxy layers route requests to the correct node

Key Takeaway: A distributed key-value store partitions data across multiple nodes to overcome single-machine limits, requiring routing logic and coordination that single-node stores don't need.