Types of Quorums

Applies To: Windows Server 2003, Windows Server 2003 R2, Windows Server 2003 with SP1, Windows Server 2003 with SP2

Standard Quorum

As mentioned above, a quorum is simply a configuration database for MSCS, and is stored in the quorum log file. A standard quorum uses a quorum log file that is located on a disk hosted on a shared storage interconnect that is accessible by all members of the cluster.

Note

It is possible to configure Server clusters to use the local hard disk on a server to store the quorum, but this is only supported for testing and development purposes, and should not be used in a production environment. Each member connects to the shared storage through some type of interconnect (for example, SCSI or Fibre Channel), with the storage consisting of either external hard disks (usually configured as RAID disks), or a storage area network (SAN), where logical slices of the SAN are presented as physical disks.

Note

It is important that the quorum uses a physical disk resource, as opposed to a disk partition, as the entire physical disk resource is moved during failover.

Standard quorums are available in Windows NT 4.0 Enterprise Edition, Windows 2000 Advanced Server, Windows 2000 Datacenter Server, Windows Server 2003 Enterprise Edition and Windows Server 2003 Datacenter Edition.

The following is a sample diagram of a standard quorum in a four-node cluster.

48bb2199-5436-4de7-ae48-caf546461c51

Figure 1: Standard quorum in a four-node cluster

Majority Node Set Quorums

A majority node set (MNS) quorum is a single quorum resource from a server cluster perspective. However, the data is actually stored by default on the system disk of each member of the cluster. The MNS resource takes care to ensure that the cluster configuration data stored on the MNS is kept consistent across the different disks.

Majority node set quorums are available in Windows Server 2003 Enterprise Edition, and Windows Server 2003 Datacenter Edition.

The following is a sample diagram of an MNS quorum in a four-node cluster.

9ba21fdc-70cb-4d7a-bbd3-8455c205880a

Figure 2: MNS quorum in a four-node cluster

While the disks that make up the MNS could in theory be disks on a shared storage fabric, the MNS implementation that is provided as part of Windows Server 2003 uses a directory on each node’s local system disk to store the quorum data. If the configuration of the cluster changes, that change is reflected across the different disks. The change is only considered to have been committed, that is, made persistent, if that change is made to:

(<Number of nodes configured in the cluster>/2) + 1

This ensures that a majority of the nodes have an up-to-date copy of the data. The cluster service itself will only start up and therefore bring resources online if a majority of the nodes configured as part of the cluster are up and running the cluster service. If there are fewer nodes, the cluster is said not to have quorum and therefore the cluster service waits (trying to restart) until more nodes try to join. Only when a majority or quorum of nodes are available, will the cluster service start up, and bring the resources online. In this way, because the up-to-date configuration is written to a majority of the nodes regardless of node failures, the cluster will always guarantee that it starts up with the latest and most up-to-date configuration.

In the case of a failure or split-brain, all partitions that do not contain a majority of nodes are terminated. This ensures that if there is a partition running that contains a majority of the nodes, it can safely start up any resources that are not running on that partition, safe in the knowledge that it can be the only partition in the cluster that is running resources (because all other partitions are terminated).

Given the differences in the way the shared disk quorum clusters behave compared to MNS quorum clusters, care must be taken when deciding which model to choose. For example, if you only have two nodes in your cluster, the MNS model is not recommended, as failure of one node will lead to failure of the entire cluster, since a majority of nodes is impossible.