sharednothing

Shared-nothing is a distributed computing architecture in which each node operates independently with its own memory, processing, and storage resources, and there is no shared disk, memory, or other central resource. Nodes communicate through message passing and coordination is achieved without centralized shared hardware. This design aims to minimize contention and allow horizontal scalability by adding more nodes.

Key characteristics include data partitioning and local processing. Data is typically distributed across nodes using sharding or partitioning schemes, so queries can be executed in parallel, often with minimal cross-node data transfer. Each node handles its own I/O and computation, and failure of one node is ideally isolated from others. Systems commonly use replication for fault tolerance and rely on distributed protocols to maintain consistency and coordination where required.

History and context: The term shared-nothing was popularized in the database community in the 1980s as a contrast to shared-disk and shared-memory architectures. It is associated with work by Michael Stonebraker and colleagues on scalable parallel databases, and it has since influenced the design of many modern distributed data stores and analytics platforms.

Applications and examples: Shared-nothing architectures underpin many large-scale databases and NoSQL stores, such as Cassandra, Riak, and MongoDB with sharding, as well as various distributed analytics and data warehousing systems. They are also used in cloud-native services and web-scale applications that require linear scalability and fault isolation.

Advantages and challenges: Benefits include improved scalability, fault isolation, and reduced resource contention. Challenges involve the complexity of distributed query processing, cross-node transactions, data rebalancing during growth, and the impact of network latency. Achieving strong consistency can require coordination that partly mitigates some performance advantages.