It seems to me that unless the system is a read-only service (in which case, full data replication can be done across every node), you cannot really have CP. If the system is partitioned due to network loss, writes to one partition cannot be seen on the other partition till the network is restored. In other words, while the network is lost, consistency is lost