A Distributed Data Store Model Satisfying Sequential Consistency or Causal Consistency with Operation Logs
Abstract
A distributed data store can satisfy two properties out of three properties which are (strict) consistency, availability and partition-tolerance. In case of distributed data stores satisfying availability and partition-tolerance, they can satisfy weak consistency, especially causal consistency, which is the strongest consistency that can cohabit with other two properties. Moreover, if any networks between nodes have no problem and very low latency, the distributed data store can satisfy stronger consistency than causal consistency. Sequential consistency is one of the stronger consistency than causal consistency. In order to satisfy sequential consistency, a distributed data store needs to equalize an order of data changing in all nodes. In this paper, we propose distributed data store model containing special nodes “casting nodes” and algorithms in order to decide an order of operations. Thanks to the casting nodes, our model can satisfy sequential consistency when all networks can connect, and our model can satisfy causal consistency when any networks disconnect.