Recommended Server Configuration | Microsoft Application Center 2000 Resource Kit 2001

Before attempting to set up a cluster controller and create a cluster, you have to assess the processing capabilities of the server that you plan to use as the controller. This server has to have enough resources to comfortably run Microsoft Windows 2000 Server and Microsoft Windows 2000 Advanced Server, Application Center, Application Center Event and Performance Logging (optional), Internet Information Services version 5.0 (IIS), and the applications it's serving—whether they are Web-based or COM+ applications.

NOTE
The Application Center Administrative client runs on any Windows 2000 operating system. Because it includes the Windows 2000 Management Console (MMC), you can run the client on a server outside a cluster and administer any server on a cluster—provided, of course, that you have the appropriate access permissions and can provide the necessary authentication information. In addition to supporting remote administration via a graphical client, Application Center provides a command-line tool that you can use for most cluster administration tasks. The command-line tool is covered in more detail in Chapter 11, "Working with the Command-Line Tool and Scripts," which includes information about using scripts and batch files.

You can use the following configuration for a server running Windows 2000 Server and IIS as a guideline for configuring a server for use on an Application Center cluster.

Memory

The official minimum memory for running Application Center on Windows 2000 Advanced Server is 256 MB of RAM on a 400 MHZ system. However, as you probably know, there are several factors to consider when determining a server's memory requirements.

Windows 2000 Server and IIS require a minimum of 256 MB of RAM; however, from 512 MB through 1 GB is recommended. The high end of the range should be considered if the site is hosting an e-commerce application, contains a large amount of content, uses dynamic pages extensively, uses COM+-based applications, or has a high volume of traffic. Remember that the IIS cache size defaults to half the available amount of real memory.

IMPORTANT
You should monitor memory and cache settings on an ongoing basis. For more information about these settings, see Chapter 8, "Creating Clusters and Deploying Applications."

Fixed Disk

The disk partition must contain adequate space for all the installed programs, paging file space, and site content. You also have to factor in the space required for content replication. Because the replication engine copies all the content to a temporary directory on the destination server, and then moves these files to the appropriate folders during synchronization, the required disk space is approximately double the volume of the content to be replicated.

NOTE
The Synchronization Service uses a two-phase commit process to ensure data integrity, which is why the replication engine uses a temporary directory on a target.

Finally, don't forget to allow enough free space to support disk defragmentation. A minimum of 15 percent of the disk should be free in order to support effective defragmentation. (A higher percentage of free disk space will improve disk defragmentation.)

Network Adapter

Each system should have at least two network adapters if it is part of a load-balanced cluster (and must have at least two, if Network Load Balancing [NLB] is to be used). The front-end adapter (also called the load-balanced adapter) is used for front-end traffic such as NLB heartbeats, convergence, and load balancing. The back-end adapter (also called the management-traffic adapter) is used for back-end traffic generated by different cluster activities, notably content replication and synchronization.

Some form of name resolution should be enabled for the back-end network adapter, and NetBIOS needs to be bound to the network adapter in order for the Application Center name resolution service to work (this is recommended, but not mandatory; see the following note). Typical options for providing DNS name resolution are:

Custom host files.
Firewalls that support independent name resolution on each side of the firewall (this "split DNS" is supported by products such as Gauntlet and Sidewinder).
Independent DNS servers.

NOTE
There are two reasons for requiring two network adapters. First, it enables NLB to bypass a loop back condition that occurs when using unicast mode. Second, the replication engine that Application Center provides uses COM calls extensively, and replication will fail if a connection gets dropped or reset. IP address changes or deletions can cause connection drops or resets, which is why using the front-end network adapter for the high volume data transfers that typify content replication is risky. This issue isn't exclusive to replication but includes other Application Center features, such as cluster services and monitoring.

IP Addresses

Application Center supports several IP address binding scenarios for the front- and back-end adapters. In most cases a single DHCP-assigned address is bound to the back-end adapter. The front-end adapter's IP address story isn't quite as straightforward.

NOTE
After installing Application Center you should verify that NetBIOS over TCP/IP is enabled for each IP address. You can access this setting by opening the Internet Protocol (TCP/IP) Properties dialog box for the IP bound to the adapter. Next, click Advanced TCP/IP Settings, and then click the WINS tab.

For the Controller

If your cluster uses NLB, the controller requires a minimum of one static IP address bound to the front-end adapter. This single IP address serves as the cluster IP address for the adapter and there is no dedicated IP address, which is restrictive from a network management perspective. For example, you can't ping the front-end adapter on a specific server by using the IP address because the address is common to all the cluster members in a load-balanced cluster. The various server and cluster configurations that we use in this book all use two IP addresses for this very reason. (For more information about load balancing, adapter configuration, and traffic implications, see Chapter 5, "Load Balancing.")

For a Member

In an NLB cluster, the front-end adapter requires a minimum of one IP address, which can either be static or assigned by DHCP. In the latter case, Application Center will automatically change the address setting from Obtain an IP address automatically to Use the following IP address. When the member is added to the cluster, Application Center binds the cluster IP address to the adapter. Once again, network management considerations should determine whether you want to use one or two IP addresses on the front-end adapter.

Figure 4.1 illustrates a server configuration for an Application Center cluster member that's using NLB. In this particular example, there are two static IP addresses bound to the front-end network adapter. The first is a dedicated static IP address that enables you to communicate directly with the front-end adapter. The second address is the cluster IP address that carries the load-balanced cluster traffic.

click to view at full size

Figure 4.1 Cluster member configuration

Subnets

Notice that the front-end and back-end adapters in Figure 4.1 are on separate subnets. There are two reasons for this. First, using separate subnets provides a more secure implementation by isolating the back-end (internal) traffic from the front-end, or external, traffic. (Another technique for isolating network traffic is by using network segments.) Second, using separate subnets improves traffic distribution over the network adapters when you're using NLB. This has to do with the way Application Center configures adapter interface metrics and the way NLB routes traffic. If you create a cluster that uses NLB, Application Center sets the interface metric for the load balanced adapter to be one higher than the interface metric for other cards on the same computer.

For example, let's assume that you have a server configured with two adapters, both of which are on the same subnet. If the interface metric for the back-end adapter is 1, Application Center sets the interface metric for the load-balanced adapter (the front-end) at 2.

When a response is sent to the client, NLB routes it to the adapter on the same subnet that has the lowest interface metric. In this scenario, outgoing traffic is routed to the back-end adapter, which is where Application Center transmits all cluster management and synchronization traffic. Depending on the amount of traffic on the back end, this could have a negative impact on services, such as the Synchronization Service.

Separate subnets aren't mandatory; however, they are supported if you decide to use them for your clusters.

Although Application Center will run on any server that meets the minimum requirements, homogenous hardware is recommended. This is the best way to ensure balanced and consistent performance across a cluster, as well as making it easier for you to tune your servers for optimum performance. If you're using NLB or one of the compatible third-party load balancers, you can adjust load-balancing weights to compensate to some extent for performance differences between servers.

However, you have to remember that each cluster member is synchronized to the controller. As a result, there is very little leeway in customizing individual configurations, especially IIS.

NOTE
Multiple disk partitions can be used, but the idea of homogeneity extends to the file system structure on a disk partition. Identical file system structures are required (System Root, Program Files path, Application Center path), and we strongly recommend the use of NTFS rather than FAT32. For more information about the security and replication issues related to drive formats, refer to Chapter 6, "Synchronization and Deployment."

The main factors in selecting homogenous systems for a cluster are the number of CPUs, CPU speed, disk partitioning/formatting, and memory on each server.