System Admin

clusterA group of two or more nodes within a system supporting clustering.<br />
<br />
Overview<br />
When a client on a network tries to access shared resources or applications on a cluster, the cluster appears to the client as a single node or server instead of the group of nodes or servers it really is.<br />
<br />
In the Cluster service of Microsoft Windows 2000 and Windows .NET Server Enterprise Server and Datacenter Server, each node in a cluster is a completely independent computer system that must be running Windows 2000 Enterprise Server. Typically, such nodes are connected by a shared storage bus such as an external Small Computer System Interface (SCSI) disk subsystem or RAID array.<br />
<br />
clustering<br />
Any technology that enables two or more servers to appear to clients as a single system.<br />
<br />
Overview<br />
A cluster consists of a group of computers functioning together as a unit, running a common set of applications, and presenting a single image to client systems. Clustering can be implemented in various ways, but its basic goals are to provide businesses with high availability, high reliability, and high scalability solutions for mission critical business operations.<br />
<br />
Clustering solutions generally come in two basic types:<br />
<br />
<br />
Stateful clustering: The goal of this type of clustering is to provide high availability and high reliability for fast, uninterrupted service in high- demand environments that can tolerate minimal downtime (stateful clustering is not designed to scale applications out to handle more users-this is the purpose of stateless clustering discussed next). Stateful clustering works by connecting independent computer systems into a single entity called a cluster, with each system within the cluster being called a node. Generally some multiple of two is used as the number of nodes within a cluster, for example, 2, 4, 8, 16, or 32 nodes. The different nodes within a cluster are usually connected using a shared disk subsystem which typically consists of a hard disk system or RAID-5 array connected to each node using a fast Small Computer System Interface (SCSI) bus or fiber channel connection. The result is a cluster of computer systems that acts and functions as if it were a single system. In stateful clustering, the nodes within a cluster generally share the workload, and when one node fails its workload fails over (transfers to) another node in the cluster with no interruption of services from the user's perspective. When the failed node comes back online, the workload fails back to this node and normal operation of the cluster resumes. Failover in stateful clustering systems can be implemented in different ways, and this is discussed below. An example of a platform supporting stateful clustering is the Cluster service of Microsoft Windows 2000 Enterprise Server and Datacenter Server. Another name for stateful clustering is shared storage or shared something clustering.<br />
<br />
<br />
Stateless clustering: The goal of this type of clustering is to provide high availability and high reliability by enabling administrators to scale out applications to meet increased demand as the number of users and traffic generated increase. Stateless clustering uses a group of nodes (servers) that are not connected in any way apart from the underlying network connectivity. No failover occurs between nodes when a node in the cluster fails. Instead, some form of load balancing is used to share the workload between the different nodes, and if one node fails the other nodes pick up the extra workload with no interruption of services. Stateless clustering is supported by three Microsoft products: Network Load Balancing (NLB), Component Load Balancing (CLB), and Application Center 2000. Another name for stateless clustering is shared- nothing clustering, and SQL Server 2000 supports a form of shared-nothing clustering known as Federated Server Groups.<br />
<br />
Stateful clustering solutions themselves generally fall into three different categories, depending on if and how failover occurs between different nodes:<br />
<br />
<br />
Active/active clustering: This type of clustering makes the most efficient use of system resources because there are no redundant nodes: all nodes run active processes. If one node of a cluster fails, other nodes take on the failed cluster's workload. The latency for failover in this scenario is typically 15 to 150 seconds, depending on the hardware/software configuration. Active/active clustering is supported by the Cluster service of Microsoft Windows 2000 Enterprise Server and Datacenter Server, and by the Cluster service of Windows .NET Server (discussed later in this article).<br />
<br />
<br />
Active/standby clustering: Nodes are paired within a cluster, with one node designated to take over should another node fail. If an active node fails, a standby node assumes its workload. Latency for failover is also 15 to 150 seconds. Active/standby clustering is a more expensive solution than active/active because the standby node is essentially doing nothing unless the active node fails.<br />
<br />
<br />
Fault-tolerant clustering: Nodes are paired within a cluster, and all nodes perform all tasks simultaneously. This is an expensive solution from a hardware point of view, but latency for failover is reduced to a second or less.<br />
<br />
Marketplace<br />
Many different clustering solutions are in the marketplace, but this article focuses on four different clustering technologies delivered by Microsoft platforms and products, namely:<br />
<br />
Windows clustering, previously known as Microsoft Cluster Services (MCSC)<br />
<br />
Network Load Balancing (NLB)<br />
<br />
Application Center 2000<br />
<br />
Component Load Balancing (CLB)<br />
<br />
You can find additional information in separate articles on each of these four solutions.<br />
<br />
Windows clustering is a feature of Microsoft Windows 2000 Advanced Server and Datacenter Server and of Windows .NET Server. Windows clustering is probably Microsoft's best-known clustering platform and was originally developed for Microsoft Windows NT Server Enterprise Edition where it was code-named Wolfpack during its development. Windows clustering is a stateful clustering solution that enables system architects to create clusters from groups of independent computer systems and to run and manage cluster-aware applications. Using Windows clustering, you can build two-way clusters (that is, clusters with only two nodes) on Windows 2000 and Windows .NET Server Enterprise Server edition or four- way clusters on Windows 2000 and Windows .NET Server Datacenter Server edition (Windows NT Server Enterprise Edition supported only two-way clustering). In Windows clustering a cluster connects nodes together using a shared file system and clusters can utilize active/active clustering for maximum reliability and availability. Windows clustering makes an excellent choice for clustering database and messaging applications for enterprises.<br />
<br />
Network Load Balancing (NLB) is a stateless clustering solution included with Windows 2000 and Windows .NET Server Enterprise Server and Datacenter Server editions, and it was formerly called Windows Load Balancing Services (WLBS) on the Windows NT Server 4 platform. NLB provides load balancing of Internet Protocol (IP) traffic to up to 32 independent network nodes (servers) and is typically used to build farms of Web servers or Exchange 2000 Outlook Web Access (OWA) servers for large enterprises. When one node in an NLB cluster goes down, the load is simply redistributed to the remaining nodes.<br />
<br />
Application Center 2000 is a part of Microsoft Corporation's .NET Server family, and is a stateless clustering platform designed to provide a single point of management for farms of Web servers. Appcenter is typically used in conjunction with NLB and CLB to provide high availibity, high reliability clustering that can scale out to large numbers of users. Appcenter manages a collection of servers in a Web farm as a single entity and can be used to create new clusters, join servers to existing cluster, remove nodes from clusters, deploy applications and application components to different nodes within a cluster, move components between nodes of a cluster, monitor the performance of a cluster, and manage load balancing of network connections to cluster nodes and COM+ components within a cluster-aware application.<br />
<br />
Component Load Balancing (CLB) is supported by all versions of Windows 2000 Server and is used to provide load balancing of COM+ objects across distributed applications deployed on up to 16 nodes (servers). CLB is a stateless clustering solution that requires no special hardware but needs Microsoft Application Center 2000 in order to operate.