The information in this article applies to:
SUMMARYThe purpose of this article is to describe how to interconnect two or more Windows NT Load Balancing Service (WLBS) servers in a switched environment. For Windows Load Balancing Service (WLBS) to work properly, every packet directed to the virtual IP (VIP) address must reach all WLBS hosts. The WLBS host then decides which packets to accept. MORE INFORMATION
When the cluster network adapter of each WLBS host is connected to a port on a Layer 2 switch (the same switch) instead of a hub, the switch tries to learn the media access control (MAC) address of the computer connected to each of its ports so it can associate a port to a MAC address. Ethernet switches send frames to a MAC address using the port associated with the MAC address. UnicastNOTE: For optimal results, the default gateway should be configured as follows; the default gateway address should be set on the dedicated network adapter. Do not set the gateway address for the WLBS Virtual adapter in the TCP/IP configuration dialog.This will result in requests to the cluster flowing through the cluster network adapter, and the appropriate host will reply through its dedicated network adapter (because this network adapter has the gateway information associated with it), which connects to a dedicated switch port with either of the Unicast configurations. Masking the WLBS Cluster MAC AddressWhen you use unicast, keep the MaskSourceMAC registry value set to its default value of 1. This forces the cluster to use a "dummy" MAC address when sending packets through the switch. The switch maps the dummy MAC address to a port, but sends traffic to the real cluster MAC address to all ports in the switch. If a switch does not have a MAC address associated to a port, it sends the frames to all ports. This is port "flooding" and can only be contained with a VLAN. This configuration has the highest bandwidth, and completely eliminates collisions.Using a HubReconfigure MaskSourceMAC = 0 on all WLBS nodes. Connect all of the cluster network adapters to a hub, which is in turn uplinked to a switch port. Connect the dedicated network adapter to a free switch port. This configuration permits the switch to learn the WLBS cluster MAC address, eliminating port flooding without the requirement to use a VLAN. This configuration has a bandwidth limitation on the hub, however this approach automatically improves the pipelining of traffic, and minimizes collisions.MulticastNOTE: When the local router needs to send a packet to the VIP, the local router uses address resolution protocol (ARP) to determine the VIP's MAC address. WLBS replies to these ARP requests. When you mask the source MAC address, the ARP response from WLBS has a dummy source MAC address in the Ethernet frame, but contains the correct cluster MAC address in the ARP header. Some routers cannot make this ARP mapping and require a static ARP entry to be made in the router. For additional information about static ARP requirements, click the article number below to view the article in the Microsoft Knowledge Base:Q197862 WLBS Cluster Is Unreachable from OutsideThe cluster uses a multicast MAC address mapped to a unicast IP address. The switch does not associate the multicast MAC addresses to a port, so the switch sends frames to this MAC address on all ports. IP Multicast pruning implementations cannot limit the port flooding and a VLAN would have to be used. Notice that multicast provides no advantage over unicast from the switches perspective, and the increased multicast processing overhead for routers and switches may lead to degraded or slower performance. You should carefully analyze the network impact when multicast is used, to avoid congesting other network devices. Additional query words: convoy
Keywords : |
|
Last Reviewed: December 30, 1999 © 2000 Microsoft Corporation. All rights reserved. Terms of Use. |