Microsoft Cluster Server allows for resources to be members of a group of
like resources. Resources may run on only one node at a time. Groups are
the basic failover units within a cluster and may be owned by only one
system at a time. If a resource failure occurs within a group, the resource
manager may try to bring the resource online or failover the entire group
to another node in the cluster. During failover, clients using cluster
resources may experience little or no change in activity, or else may need
to reconnect. Client impact depends greatly on the type of resources they
use and other resource factors.
The time required to bring a resource online may depend on the following
factors:
- Type of resource
- Resource properties
- Group properties
- Dependent resources
Some resources may reach an online state very quickly whereas others take
several seconds to go online. The amount of time required depends on the
type of resource. For example, an IP address may go online very quickly,
but a database server service may take several seconds. The database may
check integrity of data files while an IP address does not require the same
initialization procedures. A network name may take time to be registered on
the network and to check for duplicate network names.
Resources may be configured to stay offline after a specified number of
failures. They may also be restricted from running on specific nodes within
the cluster. Use Cluster Administrator to modify these properties.
Groups may have a preferred owner and may be configured to failback
automatically if the preferred owner node is available. Groups may be
configured not to failback to another node or to failback only during a
specific time window.
Resources may also be dependent on other resources within their group. As a
result, the dependent resources will not come online until the resources
they depend upon are online.
Additionally, if a node fails, surviving cluster nodes must arbitrate for
access to the quorum disk resource. Quorum disk arbitration may occur
within seconds. However, the performance of hardware and attached devices
on the shared SCSI bus may influence the speed of this process.
All of these factors and many others may influence resource startup or
failover time requirements. Hardware and network performance may also
impact cluster resources.