Recently, I communicated with a financial institution that the client is doing a distributed core transformation, adopting a three-region five-center Multi-Active five-copy scheme recommend by an O database, which is a very novel scheme, learn more about the architecture of the solution. City A has two data centers, A1 and A2 respectively. City B, 300 kilometers away from City A, also has two data centers, B1 and B2 respectively, city C, which is about 500 kilometers away from City A, will build data center C. The single-system deployment architecture A1,A2,B1,B2, and C data centers each deploy one copy, in table units, A1 is the main unit, and the other is slave nodes, forming a five-copy deployment scheme, of course, some partitions can work in B- based, and A- based is slave, thus forming an architecture of mutual master-slave. The distributed database performs master-slave replication based on a single table partition. Based on the majority theory, A1 data needs to be synchronized to any two replicas such as A2,B1,B2, and C to be returned to the application server, in other words, to synchronize logs across 300 kilometers, you need to rent a high-quality leased line bandwidth from the operator.

It turns out that the magical Multi-Active solution of three regions and five centers is like this. It needs to be replicated synchronously across cities. The bandwidth requirement must be very high. How much does it cost? In addition, there are many network element devices in the middle of the cross-city distance, with a single IO delay of at least 10ms or more. Can the service really run? What should I do if there are link jitter, freezing, and timeout databases? Wait all the time or how to isolate quickly? I don't know how the deployment effect of these three-region five-center Multi-Active solutions is? Wait a few years to see.

Source: Bi xunshuo