Clustering: the smart way to avoid disaster

Today, we’ll continue to look at the new and improved functionality in Windows Server 2003 by examining cluster services, a feature which, while not glamorous, can be a major draw when trying to get funding for the upgrade.

Cluster services, the ability for one server to take up the load instantaneously when another fails, gets the starring role in a field called disaster prevention. Disaster recovery is good, and everyone needs a disaster recovery plan – but it’s so much better if you never experience a real disaster. Mark Ehr, writing in a Network World Network/Systems Management Newsletter (https://www.nwfusion.com/newsletters/nsm/2003/0120nsm2.html) believes that downtime cost estimates of $1 million per hour might be somewhat exaggerated, but even at a tenth of that it’s still a whole lot more than the cost of clustering.

The first big change in Windows 2003 is simply the size of the allowed clusters. While Windows 2000 Enterprise Edition supports 2-node clusters, and Windows 2000 Datacenter server handles 4-node clusters the equivalent Windows 2003 servers will all support 8-node clustering. Additionally, Windows 2003’s 64-bit servers will also support clustering. This means the larger memory capacity of the 64-bit servers (up to 4 terabytes of RAM compared to Windows 2000’s 64M byte) means better load balancing in the cluster as well as faster and more efficient failover should a disaster strike.

Clustering is generally based on the use of a shared storage system, a network-attached storage device partitioned among the servers, but allowing the partitions to be remapped should a server fail. Windows 2003 clustering though, also will work when there is no shared storage – at least as we now think of it. Other techniques such as log shipping or software “disk” – even file system replication and mirroring – can be used to make date available on multiple nodes in the cluster.  A process of storage abstraction is used, perhaps best visualized as “virtual storage” which is then handed off to drivers that are particular to the storage method chosen.

Hardware manufacturers will even be able to build systems to support geographically dispersed clusters. While this functionality will be available for you the network manager, to set up, it’s highly recommended that outsourced third party specialists be used since storage and communications need to be optimized.

You need to think of clustered servers as being required for business critical applications. No matter how little you think downtime costs, it’s still more than the cost of setting up a cluster. And you will sleep better. Nighty night.