Geo Clustering exists in many options, and dependent highly on the requirements and technical capability. This post is to discuss some options and things to consider before deploying any geo-cluster.
Data GEO- Redundancy
The first dependency in clustering is storage capability. Data from the workload in the cluster will be written to disk and that data needs to be available on both sites. Within Microsoft SQL AlwaysOn can replicate the data for the instances and ensure it is available on both sites. It is also possible to have the storage perform data mirroring.
When sending data from site A to site B, two options exist: Synchronous and A-Synchronous.
Synchronous: Data is written to BOTH sites before the application or server receives a successful write notification
A-Synchonous: Data is written to the primary site, the application or server receives the write, and THEN the data is written to the second site.
Within a synchronous architecture, there is very limited chance of data-loss upon a failure, as the application knows the data is written in two locations. With A-synchronous data loss can occur.
While synchronous looks most tempting, it requires fast connections between the storage / servers in order to reduce latency for every I/O write action. Therefore this is not always possible and a-synchronous is the only option left.
Storage mirroring or AlwaysOn data replication must be used to provide data geo-redundancy