Windows 2003 cause the network switch port flood

Wlewx11 · ‎02-13-2009

My company have setup a win2k3 NLB service and almost bring down the entire switch.within 10-15 minutes all the network connection jam (port flooding).during the time occur, we can't even ping the gateway or any. After 15minutes, all the connection recover by itselt.

we r using unicast and the NLB servers directly plug into L2 switches.

Ethereal discover ,

Eg

source= Cisco_XX:XX:XX

destination=Cisco_XX:XX:XX

Protocol=LOOP

Info=Reply

source= Cisco_XX:XX:XX

destination=spanning-tree-(for-bridges)_00

Protocol=STP

Anyone can tell me it was NLB cause it?

if Yes, any better solution for Win2k3 NLB service

b.julin · ‎02-13-2009

From: http://technet.microsoft.com/en-us/library/cc783135.aspx

Q. How do I Reduce Switch Flooding Caused by Network Load Balancing?

A. For Windows 2000, the only choice is to isolate the NLB hosts. This can be done in one of two ways. First the hosts can be homed to their own LAN or Virtual LAN. This will work for either unicast or multicast modes. Second, for the multicast mode, the switch ports to which the hosts are attached can mapped to the cluster MAC address via static entries in the switchs Content-Addressable Memory (CAM) table.

Windows Server 2003 introduces a new feature, called IGMP multicast, as an additional means of limiting switch flooding. This mode can be used if the switch supports IGMP snooping. Note however, that this mode requires the cluster to be configured in multicast mode.

Wlewx11 · ‎02-15-2009

I still prefer to use unicase instead multicast.

Would like to know anyone use below step before (putting all the NLB connected to Hub then uplink back to switch)

http://support.microsoft.com/?id=193602

Set all the interface in MaskSourceMAC to 0

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\WLBS\Parameters\Interface