CONCEPT Cited by 1 source
Shared-everything architecture¶
A shared-everything architecture stores all data in a single shared storage system accessible to every compute node in the cluster. Every node can read and write every piece of data; there is no horizontal partitioning of the data layer.
Contrast with shared-nothing, where each node owns its slice of the data and cross-node requests must cross a network boundary.
Trade-offs¶
-
Simpler to implement + easier consistency. All nodes see the same data through the same storage layer; there's no data-routing layer, no consistency-across-shards problem. In the HTAP context specifically, the shared-everything class is described as "simplest to implement and ensure data consistency" (Source: sources/2026-04-21-planetscale-what-is-htap).
-
Scale-limited. The shared storage layer is a scalability bottleneck — a single disk subsystem or a single database instance can only grow so far before vertical-scaling limits dominate. Shared-everything systems tend to hit IOPS, bandwidth, or lock-contention ceilings that shared-nothing systems route around. In the HTAP context: "can be limited in scalability" (Source: sources/2026-04-21-planetscale-what-is-htap).
-
Natural home for compute-storage separation. Modern shared-everything systems often split stateless compute nodes from shared durable storage (Aurora-style, Snowflake-style) — compute can scale horizontally while data stays in one logical location.
Where it shows up¶
- Classical RDBMS single-instance deployments — Postgres, MySQL running on a single machine with local disk or SAN.
- Aurora / Neon-style compute-storage-split databases — many compute replicas share one redo-log-backed storage fleet.
- Shared-everything HTAP systems — in-memory or disk-based databases that keep a single logical data surface to avoid consistency-across-shards complexity.
Seen in¶
- sources/2026-04-21-planetscale-what-is-htap — canonical wiki introduction as one of the three HTAP architecture classes. Savannah Longoria (PlanetScale, 2023-12-01) names shared-everything as "all data is stored in a single shared storage system" with simplest-implementation + natural-consistency but scalability-limited trade-off profile.