Addendum

With NLB, the data frame is forwarded to all the servers for them to perform load-balancing.
NLB Multicast Mode Scenario
Consider a sample topology in which four servers, namely S1 through S4, are configured as a cluster or a
farm. This set of servers is connected to a Layer 3 switch, which in turn is connected to the end-clients.
They contain a single multicast MAC address (MAC-Cluster: 03-00-5E-11-11-11).
In multicast NLB mode, a static ARP configuration command is configured to associate the cluster IP
address with a multicast cluster MAC address.
With multicast NLB mode, the data is forwarded to all the servers based on the port specified the Layer 2
multicast command, which is the mac-address-table static <multicast_mac> multicast
vlan <vlan_id> output-range <port1>, <port2>
command in CONFIGURATION mode.
Limitations With Enabling NLB on Switches
The following limitations apply to switches on which you configure NLB:
• The NLB unicast mode uses switch flooding to transmit all packets to all the servers that are part of
the VLAN. When a large volume of traffic is processed, the clustering performance might be impacted
in a small way. This limitation is applicable to switches that perform unicast flooding in the software.
• The ip vlan-flooding command applies globally across the system and for all VLANs. In cases
where the ARP replies contain a discrepancy with the Ethernet SHA and ARP header SHA frames and
NLB is applicable, flooding of packets over the relevant VLAN occurs.
• The maximum number of concurrent clusters that is supported is eight.
Benefits and Working of Microsoft Clustering
Microsoft Clustering allows multiple servers using Microsoft Windows to be represented by one MAC
address and IP address in order to provide transparent failover or balancing. FTOS does not recognize
server clusters by default; it must be configured to do so. When an ARP request is sent to a server cluster,
either the active server or all of the servers send a reply, depending on the cluster configuration. If the
active server sends a reply, the Dell Force10 switch learns the active server’s MAC address. If all servers
reply, the switch registers only the last received ARP reply, and the switch learns one server’s actual MAC
address; the virtual MAC address is never learned. Because the virtual MAC address is never learned,
traffic is forwarded to only one server rather than the entire cluster, and failover and balancing are not
preserved.
To preserve failover and balancing, the switch forwards the traffic destined for the server cluster out all
member ports in the VLAN connected to the cluster. To ensure that this happens, you must configure the
command ip vlan-flooding on the Dell Force10 switch at the time that the Microsoft cluster is
configured. The server MAC address is given in the Ethernet frame header of the ARP reply, while the
virtual MAC address representing the cluster is given in the payload. Then, all traffic destined for the
cluster is flooded out of all member ports. Since all of the servers in the cluster receive traffic, failover and
balancing are preserved.
Enable and Disable VLAN Flooding
• The older ARP entries are overwritten whenever newer NLB entries are learned.
• All ARP entries learned after the feature is enabled are deleted when the feature is disabled, and RP2
triggers ARP resolution. The feature is disabled with the command no ip vlan-flooding.
180
Microsoft Network Load Balancing