the argument i have tried to make here is that eventual consistency does not need to be forced in tightly coupled environments that exist in a single data center. i understand that some forms of conflicts are inevitable if one wishes to update concurrently across a WAN.
regarding CAP - CA is achievable in a single data center (where as i argue - one cannot tolerate partitions anyway). the problem with Dynamo is that the overheads incurred in tolerating partitions are imposed on environments that do not suffer from partitions.
you are right that point in time consistency is no good for partition tolerance. i didn't mean to say that it was. all i wanted to point out was that very few, if any, commercial database deployments do concurrent writes across continents (and do synchronous replication across the same). cross data center replication is used for disaster recovery - and at best analytics - where point in time is just fine.
regarding CAP - CA is achievable in a single data center (where as i argue - one cannot tolerate partitions anyway). the problem with Dynamo is that the overheads incurred in tolerating partitions are imposed on environments that do not suffer from partitions.
you are right that point in time consistency is no good for partition tolerance. i didn't mean to say that it was. all i wanted to point out was that very few, if any, commercial database deployments do concurrent writes across continents (and do synchronous replication across the same). cross data center replication is used for disaster recovery - and at best analytics - where point in time is just fine.