Microsoft Windows Server 2003 Deployment Kit [Electronic resources] : Planning Server Deployments نسخه متنی

Overview of Planning for High Availability and Scalability

A highly available system reliably provides an acceptable level of service with minimal downtime. Downtime penalizes businesses, which can experience reduced productivity, lost sales, and lost faith from clients, partners, and customers.

By implementing recommended IT practices, you can increase the availability of key services, applications, and servers. These practices also help you minimize both planned downtime, such as for maintenance or service pack installations, and unplanned downtime, such as downtime caused by a server failure.

Additionally, the clustering technologies available in Windows Server 2003 can provide added levels of availability and scalability. A cluster is a group of independent computers that work together to provide a common set of services. If a cluster node (a computer in a cluster) fails, other nodes in the cluster assume the functions of the failed node. Clusters enhance the availability of critical applications and services. Clusters also increase the scalability of your deployment by allowing you to scale your solution over time to support additional clients or increased client demands. In most deployments, server clustersprovide availability, and Network Load Balancing clustersincrease scalability.

If you need the highest levels of availability (and if your budget can accommodate the expense), use both recommended IT practices and clustering for a truly comprehensive high availability deployment that is also scalable. If you determine that your organization needs to increase availability and scalability, this chapter can help you create an effective plan for implementing a high availability and scalability solution that meets your organization's current and future needs.

High Availability and Scalability Planning Process

Figure 6.1 illustrates the high availability design process and defines the steps you can take to ensure a deployment that meets your requirements for high availability.

Figure 6.1: Planning for High Availability and Scalability

Basic High Availability and Scalability Concepts

To begin designing a Windows Server 2003 deployment for maximum availability and scalability, familiarize yourself with these fundamental high availability and scalability concepts.

Basic High Availability Concepts

An integral part of building large-scale, mission-critical systems that your business and your users can rely on is to ensure that no single point of failure can render a server or network unavailable. There are several types of failures you must plan against to keep your system highly available.

Storage failures

There are many ways to protect against failures of individual storage devices using techniques such as Redundant Array of Independent Disks (RAID). Storage vendors provide hardware solutions that support many different types of hardware redundancy for storage devices, allowing devices as well as individual components in the storage controller itself to be exchanged without losing access to the data. Software solutions also provide similar capabilities.

Network failures

There are many components to a computer network, and there are many typical network topologies that provide highly available connectivity. All types of networks need to be considered, including client access networks and management networks. In storage area networks (SANs), failures might include the storage fabrics that link the computers to the storage units. For more information about SANs, see "Planning for Storage" and "Designing and Deploying Server Clusters" in this book.

Computer failures

Many enterprise-level server platforms provide redundancy inside the computer itself, such as through redundant power supplies and fans. Vendors also allow components such as peripheral component interconnect (PCI) cards and memory to be swapped in and out without removing the computer from service. In cases where a computer fails or needs to be taken out of service for routine maintenance or upgrades, clustering provides redundancy to enable applications or services to continue. This redundancy happens automatically in clustering, either through failover of the application (transferring client requests from one computer to another) or by having multiple instances of the same application available for client requests.

Site failures

In extreme cases, a complete site can fail due to a total power loss, a natural disaster, or other unusual occurrences. More and more businesses are recognizing the value of deploying mission-critical solutions across multiple geographically dispersed sites. For disaster tolerance, a data center's hardware, applications, and data can be duplicated at one or more geographically remote sites. If one site fails, the other sites continue offering service until the failed site is repaired. Sites can be active-active, where all sites carry some of the load, or active-passive, where one or more sites are on standby.

You can prevent most of these failures by using the following methods:

Proven IT practices. IT strategies can help your organization avoid some or all of the above failures. IT practices take on added importance when a clustering solution is not applicable to, or even possible in, your particular deployment. Whether or not you choose to deploy a clustering solution, all Windows deployments should at a minimum follow the guidelines and reference the resources listed in this chapter for fault-tolerant hardware solutions.

Clustering. This chapter introduces Windows Server 2003 clustering technologies and provides an overview of how they work. Different kinds of clusters can be used together to provide a true end-to-end high availability and scalability solution. For more information about end-to-end solutions, see "Using Clusters to Increase Availability and Scalability" later in this chapter. For complete information about designing clusters, see "Designing and Deploying Server Clusters" and "Designing Network Load Balancing" in this book.

Basic Scalability Concepts

In general deployments, scalability is the measure of how well a service or application can grow to meet increasing performance demands. When applied to clustering, scalability is the ability to incrementally add systems to an existing cluster when the overall load of the cluster exceeds the cluster's capabilities.

Scaling up

Scaling up involves increasing system resources (such as processors, memory, disks, and network adapters) to your existing hardware or replacing existing hardware with greater system resources. Scaling up is appropriate when you want to improve client response time, such as on a Network Load Balancing cluster. If the required number of users are not properly supported, adding random access memory (RAM) or central processing units (CPUs) to the server is one way to meet the demand.

Windows Server 2003 supports single or multiple CPUs that conform to the symmetric multiprocessing (SMP) standard. Using SMP, the operating system can run threads on any available processor, which makes it possible for applications to use multiple processors when additional processing power is required to increase the capability of a system. For more information about scaling up, see "Using Clusters to Increase Availability and Scalability" later in this chapter.

Scaling out

Scaling out involves adding servers to meet demand. In a server cluster, this means adding nodes to the cluster. Scaling out is also appropriate when you want to improve client response time with your servers, and when you have the hardware budget to purchase additional servers as needed. For more information about scaling out, see "Using Clusters to Increase Availability and Scalability" later in this chapter.

Testing and Pilot Deployments

Before you deploy any new solution, whether it is a fault-tolerant hardware or networking component, a software monitoring tool, or a Windows clustering solution, you should thoroughly test the solution before deploying it. After testing in an isolated lab, test the solution in a pilot deployment in which only a few users are affected, and make any necessary adjustments to the design. After you are satisfied with the pilot deployment, perform a full-scale deployment in your production environment. Depending on the number of users you have, you might want to perform your full-scale deployment in stages. After each stage, verify that your system can accommodate the increased processing load from the additional users before deploying the next group of users.

Clusters in particular need to be tested and then deployed in a pilot environment before you deploy them in a production environment. For more information about testing clusters, see "Designing and Deploying Server Clusters" and "Deploying Network Load Balancing" in this book.

For complete information about setting up a test environment for your Windows Server 2003 deployment and implementing a pilot deployment see "Designing a Test Environment" and "Designing a Pilot Project" in Planning, Testing, and Piloting Deployment Projects in this kit.