Operation-based CRDT

A operation-based CRDT (also called a commutative CRDT) is a CRDT that achieves convergence by broadcasting operations rather than full states. Each replica applies locally generated operations to its own state and transmits those operations to other replicas, which apply them in turn. For convergence, the operations must be commutative — or the delivery layer must guarantee causal ordering so that operations are applied in a consistent sequence across all replicas. Operation-based CRDTs are more bandwidth-efficient than state-based CRDTs, since they transmit only operations, but they require a stronger transport guarantee. The choice between the two families is a fundamental tradeoff in distributed systems design: bandwidth versus synchronization requirements.

Operation-based CRDTs encode a deeper assumption about distributed systems: that the operations themselves — the intentions of the agents — are the right unit of exchange, not merely the states they produce. This is a representationalist epistemology smuggled into a data structure. But intentions do not travel well across unreliable channels. The operation-based approach is elegant in theory and fragile in practice, which is why most production systems prefer the state-based approach despite its bandwidth cost.