Clustering extends scalability and reliability
One of the more interesting technologies in the enterprise server arena is clustering. Basically, the technology couples two or more symmetric multiprocessing systems or nodes into a single entity using specialized messaging and routing software.
By John H. Mayer
One of the more interesting technologies in the enterprise server arena is clustering. Basically, the technology couples two or more symmetric multiprocessing systems or nodes into a single entity using specialized messaging and routing software.
In a cluster, each system runs its own copy of the operating system and the application. The main difference between clustered systems and large so-called massively parallel processing servers is that, in a cluster, nodes share a common pool of storage devices.
Clusters present an additional level of complexity compared to an SMP server because each copy of the application must be aware of the others running and must lock out the other systems before updating the cluster's common pool of storage or database. But clusters offer many of the same benefits of SMP, including low cost.'
One reason clusters are so intriguing is they can serve two purposes. They can improve system availability by having one node act as a backup for the other. In this configuration, if one node fails, users are automatically switched to the second node and allowed to continue their work uninterrupted.
Alternately, servers can be can be used to scale performance beyond the limits of an individual multiprocessor system. Large applications that typically slow down a four-way server, for example, can be partitioned and spread across multiple systems in a cluster to perform a job much more quickly than any individual four-processor SMP could.
While clusters theoretically offer a way for information technology managers to circumvent the processing limitations of individual SMP servers, they present tremendous management challenges, particularly as vendors move beyond running only two systems under Microsoft Windows NT. Windows NT clusters have primarily been limited to two systems and used as a high-availability option.
Recently, IBM Corp. announced a new clustering technology called Cornhusker that will allow IT managers to string together up to eight servers in a cluster. The technology uses Microsoft's standard clustering application programming interface so any third-party software vendor writing applications to run under the API will be able to take advantage of the technology. Cornhusker is compatible with Windows NT Server 4.0, Enterprise Edition and meets current compatibility standards for Windows 2000 Datacenter Server.
NEXT STORY: Scalable NT Servers