1: \begin{abstract}

2: CRDTs are distributed data types that make eventual consistency of a

3: distributed object possible and non ad-hoc. Specifically, state-based CRDTs

4: ensure convergence  

5: through disseminating the entire state, that may be large, and merging it to other

6: replicas; whereas operation-based CRDTs disseminate operations (i.e., small states)

7: assuming an exactly-once reliable dissemination layer.

8: We introduce \emph{Delta State Conflict-Free

9: Replicated Datatypes} (\dsrdt) 

10: that can achieve the best of both worlds: 

11: small messages with an incremental nature,

12: as in operation-based CRDTs, 

13: disseminated over unreliable communication channels, 

14: as in traditional state-based CRDTs.  

15: This is achieved by defining \emph{$\delta$-mutators} to return a 

16: \emph{delta-state}, typically with a 

17: much smaller size than the full state, that is joined to both: local and remote states.

18: We introduce the \dsrdt framework, and we explain it through establishing 

19: a correspondence to current state-based CRDTs. 

20: In addition, we present an anti-entropy algorithm 

21: that ensures causal consistency,  

22: and we introduce two \dsrdt specifications of well-known replicated datatypes.

23: 

24: \end{abstract}

25: