Skip to content

CONCEPT Cited by 1 source

Shared-nothing architecture

A shared-nothing architecture gives each node in the cluster its own private storage; no two nodes share a physical storage resource. Data is partitioned across nodes and queries route to the owning node (or fan out to many nodes and merge results).

Contrast with shared-everything, where all nodes see all data through a single shared storage layer.

This is the architectural shape of horizontally-sharded systems (concepts/horizontal-sharding): Vitess, Cassandra, BigTable, Spanner shards, most OLAP warehouses' MPP compute layer, and modern HTAP variants that chose scalability over ease-of-implementation.

Trade-offs

  • Horizontally scalable. Each node's work is independent of every other node for the data it owns. Adding nodes linearly expands capacity (in the happy case). In the HTAP context: "more scalable, with possibility to scale horizontally" (Source: sources/2026-04-21-planetscale-what-is-htap).

  • Harder to implement + manage. Distributed data requires a routing layer (proxy, query planner, client library) to know which shard owns a row; cross-shard reads require scatter-gather

  • merge; cross-shard writes need two-phase commit or saga-style coordination; rebalancing data as shards grow is a first-class operational concern. In the HTAP context: "can be more difficult to implement and manage" (Source: sources/2026-04-21-planetscale-what-is-htap).

  • Failure-domain isolation is a first-order benefit. A shard's outage caps blast radius to the customers whose data lives there — see concepts/sharded-failure-domain-isolation. Shared- everything systems don't get this for free.

  • Consistency-across-shards is a distributed-systems problem. Single-shard transactions are local; multi-shard transactions need consensus (concepts/two-phase-completion-protocol) or weakened semantics (eventual consistency, read-your-writes).

Where it shows up

  • systems/vitess — MySQL-protocol sharding substrate; canonical shared-nothing OLTP system.
  • OLAP MPP warehouses — Snowflake virtual warehouses, Redshift, BigQuery execution layer.
  • Shared-nothing HTAP systems — distributed HTAP variants that chose horizontal scale over consistency-simplicity.
  • Most cloud-native NoSQL — Cassandra, DynamoDB, HBase, MongoDB sharded clusters.

Seen in

Last updated · 470 distilled / 1,213 read