>Beating the CAP theorem inside the data center is fairly easy in practice by using a 2-tiered system.
>Have a set of lightweight controller processes which use a consistent majority voting algorithm (Paxos, Consistency, Partition Tolerance) to replicate the state of the cluster. Since it's lightweight, you can run 5 or 7 of them, meaning 2 or 3 can die. This is more than enough Availability in practice, per customer input.
it isn't beating of the CAP theorem. It is decreasing the Partition Tolerance and thus getting more Availability while preserving Consistency in full accordance with CAP.
>Have a set of lightweight controller processes which use a consistent majority voting algorithm (Paxos, Consistency, Partition Tolerance) to replicate the state of the cluster. Since it's lightweight, you can run 5 or 7 of them, meaning 2 or 3 can die. This is more than enough Availability in practice, per customer input.
it isn't beating of the CAP theorem. It is decreasing the Partition Tolerance and thus getting more Availability while preserving Consistency in full accordance with CAP.