A special case of the multimaster replicated database approach is when we limit the system to exactly two nodes with a low latency interconnect (like local Ethernet). This is master-master replication, and the common distributed consistency protocols such as 2PC and 3PC don't pose an unreasonable overhead. Even over a wide area network, there is no advantage to using newer, less tried, and more complicated replication techniques that are emerging in the multinode multimaster replication environments. So, in two-node systems, 2PC is "how replication is done." Many commercial databases support database replication via the two-phase commit method. Although this is really useful and fundamental in highly available systems design, it certainly doesn't add any scalability to a solution. Given this, we won't dig much deeper. |