Configuring Network Load Balancing | Microsoft Windows Media Resource Kit (Pro-Resource Kit)

In the test scenario, NLB Manager will run on the encoding computer to manage the server cluster. You could also manage from the client computer, but that computer will be very busy during the tests. You would not be able to manage the cluster from one of the server computers because this test will use load balancing in unicast mode. In this mode, you cannot access cluster hosts from one of the servers in the cluster because the media access control (MAC) address for each host is the same. In production, the administration NIC will be used to manage the cluster, and this will enable technicians to supply separate MAC addresses to the servers for administrative access. For more information on NLB, see the sidebar.

Network Load Balancing

NLB is one of the two clustering services offered by Windows Server 2003. Available in all editions of the Windows Server 2003 family of operating systems, this service routes incoming Internet Protocol (IP) traffic across clusters of servers. NLB enhances both the availability and scalability of Internet server-based programs such as Web servers, streaming media servers, and Terminal Services. By acting as the load-balancing infrastructure and providing control information to management applications built on top of Windows Management Instrumentation (WMI), NLB can seamlessly integrate into existing Web server farm infrastructures.

NLB features include the following:

NLB Manager. Can be used as a single point of configuration and management for NLB clusters.
Virtual clusters. Can be used to configure hosts in a cluster with different cluster IP addresses.
Support for multiple NICs. Can be used to bind NLB to multiple network cards.
Bi-directional affinity. Commonly used to cluster Internet Security and Acceleration (ISA) servers for proxy and firewall load balancing.
Internet Group Management Protocol (IGMP) support. Can be used to limit switch flooding when NLB is configured in multicast mode.

For more information about the Network Load Balancing feature in Microsoft Windows Server 2003, see the product documentation.

Creating the Cluster

NLB is automatically installed and running on the servers. The Fabrikam technicians simply configure IP addresses on the 1000Base-T network adapter and the servers will start using load balancing. NLB Manager lets the technicians create, configure, and monitor the load balancing clusters remotely from the encoding computer.

The technicians follow these steps to configure and start the server cluster.

To start NLB Manager, on the encoding computer, open Administrative Tools, and then click Network Load Balancing Manager.
In the console tree, right-click Network Load Balancing Clusters, and then click New Cluster. The Cluster Parameters wizard starts.

For complete details about configuring clusters, see the NLB Manager Help. The following steps will describe the testing parameters set up by the Fabrikam technicians.
In the wizard, enter the following information:
- For Cluster IP configuration, type the static IP address for the cluster (10.0.0.5), a subnet mask (255.255.0.0), and full Internet name (WMCluster). These are placeholder values for the test LAN. When they take the servers online, the technicians will need to change this information to work with the Fabrikam network.
- For Cluster operation mode, select Unicast. Multicast with IGMP multicast support is not required for load balancing the media servers. This selection does not affect unicast and multicast as it relates to streaming Windows Media.
- Connect to one host that is to be part of the new cluster, and select the interface. In this scenario, the technicians type 10.0.0.2, click Connect, and then select the 1000Base-T interface.
A new cluster is added to NLB Manager. Next, the technicians will add the remaining two servers.
Select the cluster in the console tree, and start the wizard to add a new host. In Connect to one host that is to be part of the new cluster and select the interface, type the IP address of the next server, click Connect, and then select the 1000Base-T interface.
Repeat step 3 for the remaining server.

After you make a configuration change in the manager, the NLB service must make the changes on the servers and confirm that there are no errors. This process can take several seconds. Also, the service must perform a convergence operation whenever the state of the cluster changes, such as when a host is added, removed, or reconfigured. During convergence, the hosts exchange messages to determine a new, consistent state.

When intranet clients connect to the cluster address (WMCluster), NLB redirects them to one of the three servers in order to balance the load on the servers. Clients can also connect to an individual server by accessing its dedicated IP address, such as 10.0.0.2 (LAN01). When clients connect in this way, load balancing is bypassed.

To receive a multicast stream, clients do not connect directly to a server, so it would not make sense to use all three computers in the cluster for this purpose. In fact, there is no advantage to using multiple servers for multicasting, and it would only add unnecessary complexity to the system. When hosting a multicast broadcast, one server will be used to handle the multicast stream, such as LAN01. Users will connect to WMCluster for unicast streams, and the multicast group streaming from LAN01 for multicast.

Network Load Balancing Considerations

The NLB service can help balance server load and can provide fault tolerance when a server goes off line, but keep the following points in mind when using NLB with Windows Media Services.

Fast Reconnect

This is a new feature of Windows Media Services 9 Series that enables clients to reconnect to a server after a temporary network outage and resume streaming from the point at which the outage occurred. Fast Reconnect works by using state information held in temporary storage on the server. Therefore, in order for the client to resume a stream, it must reconnect to the server that holds the state information for the interrupted connection. In an NLB setup, however, the client could be sent to a different server when it attempts to reconnect.

This problem is handled by changing the client affinity setting in the port rules for a host. Open the properties dialog box for each host in the NLB cluster, locate the port rules for the host, and then change the Affinity option to Single. All traffic from a particular client will then be routed to the same cluster host.

For testing, however, leave Affinity set to None. Otherwise Windows Media Load Simulator, which is running on a single client, will connect to only one of the NLB servers. If that happens, you won’t be able to test the entire cluster.

Server Applications

NLB effectively redirects client requests when a server is no longer available. However, the NLB service is not aware of other server applications or the content on a server. Even though Windows Media Services may be stopped, for example, NLB will continue to redirect clients to the server. Therefore, NLB does not replace using server rollover in Windows Media metafiles.

Also, content must be mirrored properly across the cluster. All servers must contain the same digital media files, file structure, and publishing points. If the cluster is used for a broadcast using unicast streaming, the broadcast publishing points on all servers must be running and configured identically. You can automate the mirroring of content with Distribute File System (DFS), which is included with Windows Server 2003.

The Fabrikam scenario uses the NLB service, which is software load-balancing functionality included with Windows Server 2003. Several hardware-based products are also available from companies such as Cisco and Foundry.