In every failover cluster, a single resource is designated as the quorum resource. The Cluster service uses the quorum resource to perform two fundamental tasks:
- The quorum resource allows the cluster to maintain its state and configuration independently of individual node failures. The quorum resource stores a constantly-updated version of the cluster database and is accessible to all cluster nodes. When a node forms a cluster, it compares its local copy of the cluster database with the data stored on the quorum resource and updates its copy if necessary.
- The quorum resource enforces cluster unity, preventing “split-brain syndrome” in cases where communication failures result in two or more sets of nodes that can communicate within each set but not across sets. In such cases, only the node that controls the quorum resource, and nodes that can communicate with that node, are allowed to form a cluster.
A cluster uses one of two types of quorum resources, and depending on which type it uses, the cluster is either a quorum-device cluster or a majority-of-nodes cluster. The type of quorum resource, and therefore the type of the cluster, can be changed after the cluster is set up.
A quorum-device cluster requires a storage-class resource on a shared SCSI or FibreChannel bus. By default, failover clusters allow only physical disk resources to operate as quorum devices. However, third-party developers can create their own resource types to support other kinds of storage devices as quorum devices. For more information, see Creating Resource Types.
A majority-of-nodes cluster uses a set of local disks called quorum disk set as its storage media. Each cluster node contributes one local disk to the quorum disk set. Thus, while the majority-of-nodes resource behaves like any other cluster resource and allows only one owner at a time, the owner of the resource must coordinate reads and writes with the other cluster nodes. In order to establish quorum and form a cluster, a majority of the nodes (M) must be available. In other words, a majority-of-nodes cluster can form and continue to run as long as there are M >= (N + 1) / 2 active nodes (where N is the total number of nodes configured as members of the cluster). Note that there are ways to force quorum in cases where the number of active nodes is less than M.
There are advantages and disadvantages to both types of clusters. Majority-of-nodes clusters allow the creation of “geoclusters,” or clusters of nodes separated by great distances, but their failure semantics are much more restrictive than quorum device clusters. For example, without forcing quorum, a single-node majority-of-nodes cluster is not possible. Quorum-device clusters can tolerate more node failures but are geographically restricted to the length of the shared SCSI or FibreChannel bus.
Quorum Resource in MSCS?
Quorum Resource was a resource used in Microsoft clustering services – Microsoft Cluster Server (MSCS) for Microsoft Windows NT Server 4.0, Enterprise Edition, and Windows Clustering for Windows 2000 – that guarantees that data necessary for recovery is maintained consistently among all members of a cluster. The quorum resource must be a physical disk and can be owned by only one node in a cluster.
The quorum resource stores the cluster log, which is maintained by the clustering software. If a node fails and then recovers, the quorum resource is responsible for updating that node’s configuration information so that it matches all other nodes. If the nodes in a cluster cannot communicate with each other, the quorum resource determines which node can continue operating.
If the node owning the quorum resource fails, another node takes ownership of the resource. Each cluster can have only one quorum resource.