NoSQL Types Overview (Key-Value, Document, Column-Family, Graph)

NoSQL databases are classified into four distinct architectural families based on their underlying data structure and storage engine design:

1. Key-Value Stores (Redis, Memcached, DynamoDB, Aerospike):

   • The simplest model: maps a unique string key to an arbitrary value (string, hash, list, set, or binary blob).
   • Provides O(1) constant-time point lookups.
   • Best For: In-memory caching, user session management, shopping carts, rate limiting, and real-time leaderboards.

2. Document Databases (MongoDB, Couchbase, AWS DocumentDB, Firebase Firestore):

   • Stores data as semi-structured, self-describing JSON or BSON documents.
   • Documents can have polymorphic schemas (different documents in the same collection can have different fields).
   • Supports secondary B-Tree indexes on deeply nested JSON fields (e.g., items.price).
   • Best For: E-commerce product catalogs, user profiles, blogs, and content management systems (CMS).

3. Column-Family / Wide-Column Stores (Apache Cassandra, ScyllaDB, Google BigTable, HBase):

   • Organizes data into rows containing dynamic column families indexed by a Partition Key and sorted by a Clustering Column.
   • Uses Log-Structured Merge (LSM) Trees and immutable SSTables to deliver massive write throughput (hundreds of thousands of writes per second per node).
   • Best For: IoT telemetry streams, financial time-series metrics, chat history, and audit logging.

4. Graph Databases (Neo4j, Amazon Neptune, Memgraph):

   • Uses graph theory structures consisting of Nodes (Entities) connected by directed, typed Edges (Relationships) with key-value properties.
   • Implements Index-Free Adjacency: Each node holds direct physical memory pointers to its adjacent neighbor nodes, executing multi-hop recursive queries in O(1) pointer-dereference time per hop (compared to relational databases where multi-hop queries require expensive O(N log N) recursive joins).
   • Best For: Social networks ("Friends of Friends"), knowledge graphs, anti-money laundering fraud detection, and recommendation engines.