Skip to content

Database Paradigms

DB paradigms

Relational DB

  • Rigid structure enforced by table's schema
  • Has support for transactions (ACID compliant)
  • Difficult to scale horizontally (scale-up instead of scale-out)
  • Slower read operations on large analytical workloads

  • On-line Transaction Processing (OLTP)

  • Implementations

  • MySQL / MariaDB
  • Postgres (most popular open-source choice in 2026)
  • SQL Server
  • RDS & Aurora (AWS)

NewSQL / Distributed SQL

  • Combines relational guarantees (ACID, SQL) with horizontal scalability
  • Designed for geo-distributed deployments
  • Handles high write throughput without sacrificing consistency

  • Bridges the gap between traditional RDBMS and NoSQL scalability

  • Implementations

  • CockroachDB
  • TiDB
  • Google Spanner / AlloyDB
  • PlanetScale (MySQL-compatible)
  • Neon (serverless Postgres)

Key-Value DB

  • It's a large scale hash table
  • Data commonly stored in RAM (or optionally persisted to disk)
  • Perfect use-case for caches, sessions, pub/sub, and counters

  • Some implementations now also support vector search and streaming

  • Implementations

  • Redis / Redis Stack (with vector & search modules)
  • Valkey (open-source Redis fork maintained by Linux Foundation)
  • Memcached
  • Etcd (distributed config & leader election)
  • DragonflyDB (Redis-compatible, high throughput)

Document Oriented DB

  • Each document is a container for key-value pairs (typically JSON/BSON)
  • No schema (flexible/schemaless)
  • Documents are grouped together in collections

  • Good for hierarchical or nested data

  • Implementations

  • MongoDB
  • Firestore (GCP)
  • DynamoDB (AWS, also key-value hybrid)
  • CouchDB
  • SurrealDB (multi-model with document support)

Columnar / OLAP DB

  • Data is stored column-by-column instead of row-by-row
  • Optimized for analytical queries (aggregations, scans) over large datasets
  • On-line Analytical Processing (OLAP)

  • Often used in data warehousing and business intelligence

  • Implementations

  • DuckDB (in-process, embeddable — very popular for local analytics)
  • ClickHouse (real-time analytics at scale)
  • Apache Parquet + Iceberg (open table format)
  • Snowflake
  • BigQuery (GCP)
  • Redshift (AWS)
  • Apache Druid (real-time ingestion)

Wide-Column DB

  • Keys store multiple columns (values) organized in column families
  • No schema
  • Good for time-series data, historical records, high-write, low-read workloads

  • Scales horizontally to petabytes

  • Implementations

  • Cassandra
  • HBase
  • ScyllaDB (Cassandra-compatible, higher throughput)
  • Google Bigtable

Time-Series DB

  • Optimized for sequential, timestamped data points
  • Efficient compression and retention policies
  • Common in observability, IoT, metrics, and financial tick data

  • Supports downsampling and time-based aggregations natively

  • Implementations

  • InfluxDB
  • TimescaleDB (Postgres extension)
  • Prometheus (metrics, pull-based)
  • VictoriaMetrics (Prometheus-compatible, high cardinality)
  • QuestDB (SQL-based, high ingestion rate)

Vector DB

  • Connects data that is similar (semantic proximity in embedding space)
  • Stores high-dimensional vector embeddings (from ML models)
  • Enables semantic/similarity search via Approximate Nearest Neighbor (ANN)
  • Core infrastructure for RAG (Retrieval-Augmented Generation) pipelines and AI applications

  • Rapidly growing paradigm since 2023

  • Implementations

  • Pinecone: SaaS, very population for AI apps
  • Weaviate: Open-source, graphQL API
  • Qdrant: Open-source, Rust-based
  • Milvus: Open-source, large-scale
  • Chroma: lightweight, local-first, prototyping
  • Pgvector:Postgres extension
  • OpenSearch: general-purpose search engine with k-NN plugin for ANN vector search; supports hybrid (keyword + vector) queries

Graph DB

  • Connects data that is connected (explicit relationships between entities)
  • Data is represented as nodes (vertices) with properties
  • Relationships are represented as edges (can be directed and weighted)

  • Two main models: Property Graphs (Neo4j) and RDF/Triple Stores (SPARQL)

  • Use cases

  • Fraud detection in finance
  • Recommendation engines
  • Knowledge graphs

  • Social network analysis

  • Implementations

  • Neo4j
  • Amazon Neptune
  • DGraph
  • FalkorDB (Redis-based, fast property graph)
  • Kuzu (embeddable, DuckDB-style for graphs)

Search DB

  • Analyze and index text content for full-text search
  • Support relevance ranking, faceting, typo-tolerance

  • Good for search bars, e-commerce product search, log search

  • Implementations

  • Elasticsearch / OpenSearch
  • Apache Solr (built on Lucene)
  • Algolia (managed SaaS)
  • MeiliSearch (open-source, easy to self-host)
  • Typesense (open-source, fast)