A cluster is a group of computers working together as if it were a single computer. This allows access to and management of the cluster as a single unit. Each system within a cluster is referred to as a node, and all nodes have their individual NetBIOS names. Each cluster also has a NetBIOS name, which is the cluster name.
Each node within a cluster has an IP address as does the cluster as a whole. Messages sent to a specific cluster node go directly to that server, but those sent to the cluster name or cluster IP address can be responded to by with node that is online. The network clients do not need to know which node is online to be able to access the cluster, as long as they use the cluster name instead of the node name.
The private network is dedicated to communications between cluster nodes. This connection is not used for client access to the cluster nodes. Instead, public connections are used for client access to the cluster nodes.
There are four different types of clusters in the Microsoft world:
- Failover cluster: This is a cluster made up of two ore more nodes. One server is active at a time. If one server fails, the other will assume the responsibilities of the first server.
- Network Load Balancing: Allows certain services to be accessed on up to 32 services as one TCP/IP virtual server, so that access is load-balanced across multiple servers, An example of NLB would be multiple Web servers with the same content that are accessed in a way that equals the load on all the servers to provide better performance.
- Component Load Balancing: Used to distribute application services among several servers. This allows for load balancing of access to the services and for fault tolerance of the application as a while. This is a function of Application Center 2000.
- Distributed partition views: Used with SQL to distribute a single database across several servers for load balancing. For example, if there are three servers, the database could be split into thirds and each server would manage one third of the database. Because one third of the database reside on each server, client access would be balanced among them