By Mike Talon
In recent columns, we've looked at Microsoft's growing interest in disaster recovery issues in the enterprise, and I've discussed how Windows Server 2003's features specifically address DR planning. One of the most logical technologies to look at is the enhanced version of Microsoft Cluster Service (MSCS) that ships with new versions of Windows.
Clustering has been a part of Microsoft's server products since Windows NT 4, and it found some stability and flexibility in Windows 2000. Prior to Windows 2003 Server, however, clustering services had significant restrictions and drawbacks.
Most notably, in Windows NT 4, MSCS required installing the clustering at the same time as the operating system, and you couldn't install it afterward. Windows 2000 solved this issue but limited it to two nodes per cluster unless you were willing to shell out big bucks for the Datacenter Edition.
MSCS in Windows 2003 addresses these issues, and it creates much greater flexibility and scalability than was available in previous versions of the clustering system. Installation of MSCS is now automatic; installing any of the versions of the OS capable of running it also installs MSCS by default. You must specifically configure the clustering system, but you no longer need to worry about reinstalling the OS if you decide to use it later.
In addition, both versions of the OS that support MSCS can handle up to eight nodes per cluster—a significant improvement over previous versions. You can configure these multinode clusters to direct failover to specific nodes, offer multiple paths for redundancy, and effectively load balance across a greater number of machines.
Microsoft also refers to network load balancing (NLB) as a form of clustering, where multiple file servers act in concert to handle requests for files from multiple clients. While this is technically a form of clustering, most IT pros refer to MSCS clusters as "clustering," which involves configuring multiple servers to host instances of applications and take over for each other in the event of failure.
MSCS clusters can handle file services, but for the most part, NLB "clusters" can't run application instances—an important distinction. Windows Server 2003 Web Edition and Standard Edition support only NLB. Enterprise Edition and Datacenter Edition both support NLB and up to eight-node clustering.
In addition to extending the number of nodes that a cluster can contain, MSCS also allows you to place cluster nodes regardless of connections to physical disk resources. Prior to Windows Server 2003, you must connect all nodes of a cluster to the same physical disk device, making stretching a cluster to different physical locations nearly impossible.
With Windows 2003, you no longer need to attach servers to the same physical disk resource to handle cluster operations. As long as all nodes are on the same logical subnet, and you have a relatively clear line with low latency between sites, you can stretch clusters to multiple locations and still have more than one node at each site for local failover.
A word of warning: While Microsoft implemented a methodology to handle distributing the vital quorum resource between nodes, it does not have a methodology to replicate data between nodes. You'll have to either perform regular backups and restores (which is a near logistical impossibility) or use a replication tool to keep all the nodes up to date. Microsoft recommends replication tools, either hardware or host-based, to keep everything in sync.
MSCS in Windows Server 2003 has made some significant strides in its flexibility and scalability for the modern business environment. It can now create a much more reliable platform for redundancy and availability than its predecessors, and it opens up levels of protection formerly only found in the highest-priced versions of the OS.
Mike Talon is an IT consultant and freelance journalist who has worked for both traditional businesses and dot-com startups.