Although clustering of servers has been around for a long time in
Windows (dating back to Windows NT 4.0, when it was available, but
really didn’t work), clustering in Windows Server 2012 now not only
works, but also provides a series of significant improvements. In
Windows Server 2012, you can set up clustering without a lot of the
complexity that was previously necessary to get high availability
working in Windows; even the previous requirement of having a
storage-area network (SAN) with cluster share volumes (CSV) is no
longer. All these improvements in Windows Server 2012 make continuous
availability a viable solution for enterprise-class high availability.
No Single Point of Failure in Clustering
Clustering, by definition, should provide
redundancy and high availability of server systems. However, in earlier
versions of Windows clustering, a “quorum drive” was required for the
cluster systems to connect to as the point of validation for cluster
operations. If at any point the quorum drive failed, the cluster would
not be able to fail over from one system to another. Windows Server
2008 / 2008 R2 and Windows Server 2012 clustering removed this
requirement of a static quorum drive. Two major technologies facilitate
this elimination of a single or central point of failure:
majority-based cluster membership verification and witness-based quorum
validation.
The majority-based cluster membership enables
the IT administrator to define what devices in the cluster get a vote
to determine whether a cluster node is in a failed state and so needs
to fail over to another node. Instead of assuming that the disk will
always be available, as in the previous quorum disk model, now nodes of
the cluster and shared storage devices participate in the new enhanced
quorum model in Windows Server 2012. Effectively, Windows Server 2012
server clusters have better information to determine whether it is
appropriate to fail over a cluster in the event of a system or device
failure.
The witness-based quorum eliminates the
single quorum disk from the cluster operation validation model.
Instead, a completely separate node or file share can be set as the
file share witness. In the case of a GeoCluster, where cluster nodes
are in completely different locations, the ability to place the file
share in a third site and even enable that file share to serve as the
witness for multiple clusters becomes a benefit for organizations with
distributed datacenters and also provides more resiliency in the
cluster operation’s components.
Stretched Clusters
Windows Server 2012 also supports the concept
of stretched clusters to provide better server and site server
redundancy. Effectively, Microsoft has eliminated the need to have
cluster servers remain on the same subnet, as has been the case in
Windows clustering in the past. Although organizations have used
virtual local-area networks (VLANs) to stretch a subnet across multiple
locations, this was not always easy to do, and, in many cases, was not
the right thing to do technologically in IP networking design.
By allowing
cluster nodes to reside on different subnets, plus with the addition of
a configurable heartbeat timeout, clusters can now be set up in ways
that match an organization’s disaster failover and recovery strategy.
64-Node Clusters
Clustering provides redundancy of
nodes in the cluster, but now with Windows Server 2012, an organization
can have 64 nodes in the cluster, a significant improvement over the
16-node limit in earlier versions of Windows. With 64 nodes in a
cluster, an organization can set up cluster nodes for high
availability, for redundancy, and to serve as standby servers in the
event of a server failure in the cluster. With potentially 64 nodes in
the cluster, the organization can greatly improve high availability and
continuous operations of the network systems.
