NewSQL Distributed SQL
NewSQL, also called distributed SQL, aims to give the horizontal scale of a NoSQL system while keeping the SQL interface and strong ACID transactions that relational databases provide. Earlier scaling forced a choice between scale and transactions; NewSQL tries to deliver both.
How it scales
Data is split into ranges spread across many nodes, and each range is replicated for fault tolerance. Replication uses a consensus protocol like Raft, where a majority of replicas must agree on each write. This keeps copies consistent even if some nodes fail.
- Sharding spreads data so capacity grows by adding nodes.
- Consensus keeps replicas in agreement and survives failures.
- A transaction coordinator drives writes that span ranges.
Distributed transactions
A transaction touching several ranges uses a protocol such as two phase commit to make all parts commit together or none at all. Some systems use synchronized clocks to order transactions globally so reads see a consistent snapshot. The cost is added latency from cross node coordination.
Key idea
NewSQL spreads relational data across nodes with consensus replication and distributed transactions, keeping SQL and ACID at the cost of coordination latency.