There are lots of existing systems, but none of them are satisfying. Amazon's Dynamo[3] only provides eventual consistency, which is too weak for the programmer to build on to provide other services. The inconsistent views of the updates can escape and poison those other services. On the other hand, strong-consistency (linearizability) is impossible due to CAP. Bayou is probably the closest work, but it does not scale well because they achieved causal+ via log-based replay. Log-exchange-based serialization inhibits replica scalability, as it relies on a single serialization point in each replica to establish ordering.