cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
2396
Views
10
Helpful
5
Replies

Microsoft NLB on Nexus 5596T

rramlal
Level 1
Level 1

  Hi guys,

We recently installed two 5596T in a cluster at a customer. They are currently migrating their servers to a VMWARE solution and have asked if the nexus supports microsoft NLB on multicast mode.

I reached to Cisco TAC on this however I havent gotten any confirmation on the commands that are required. Can you advise what commands are required for this to allow servers to see the NLB server. So far the customer are able to migrate and the solution is working however is it recommended to set static maps on the nexus for the nlb server?

Thanks much.

5 Replies 5

richbarb
Cisco Employee
Cisco Employee

Hi,

Look this document, I applied this configuration once and worked fine.

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------

There are 3 modes to Microsoft Network Load Balancing (NLB)

1.      1.Unicast

2.      2.Multicast

3.      3.IGMP multicast  (check the IGMP checkbox in the GUI while in multicast mode)

In general,every mode uses a different sending and receiving mac address while keeping the unicast virtual IP address (VIP) constantacross all 3 modes. This concept makes switches flood traffic at layer 2 since the switch either never sees the destination mac address come in on any of its ports(and hence can’t learn it) or the multicast mac address floods. Either multicast mode, IGMP or normal multicast, also requires static ARP entries on the gateway router since Cisco routers will not learn an ARP reply with a multicast mac address tied to a unicast ip address.

Mac addresses in the 3 modes breakdown into the following components:

  • The     first number in the mac address is the type of NLB configuration: 01=IGMP,     02=Unicast, 03=Multicast (Note: bit 2 is the administered locally     multicast space)
  • The second number, (bf) is the same for unicast and multicast mode (not IGMP multicast mode     which uses the standard 01-00-5e mac address)
  • The last two (IGMP multicast mode) or four (unicast or     multicast mode) numbers are the virtual IP address, i.e. c0=192, a8=168,     04=4, 0a=10 and thus the IP of 192.168.4.10 has a multicast mac address 03-BF-C0-a8-04-0a     while an IGMP multicast mac address would be 01-00-5e-7f-04-0a

Summary of configuration

NLB mode

Switch  configuration

Router  configuration

Unicast

Mac  address-table static 02bf.xxxx.xxxx vlan y interface  

Not  required – unicast mac address with unicast ip address

Multicast

Mac-address-table static 03bf.xxxx.xxxx  vlan y interface   

n7k[5.2(1)]: mac address-table  multicast 03bf.xxxx.xxxx vlan y interface

Arp  03-bh-xx-xx-xx-xx arpa

IGMP multicast

Mac address-table static 01005exx.xxxx  vlan y interface   

Arp  01-00-5e-7f-xx-xx arpa

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Basicly you have to add the mac address static to all ports that you learned the mac address of your cluster microsoft (including the vpc peer-link if exist).

In the router (could be the N5K, if it doing the L3 boundary) you have to add the arp entry in the vlan sub-interface configuraton mode.

The cluster comes up instantly.

Richard

Thanks for the response. The server is actually virtualised and it is working now since they are directly connected to a 3com switch that uplinks to the nexus 5k.

This is temporary as they would be moving to the 10gb links that are directly connected to the 5k. The server network is flat and the L3 interface is the firewall not the n5k. So with this in mind should i statically add the mac address static to the 10GB interfaces alone since the microsoft server traffic will be coming from that link. Since that link will be a trunk uplink for the vmware cluster, how will this be affected. Please advise?

No problem at all, in the N5K you have to add all interfaces possible may to learn the mac address virtual from the cluster, even it is a trunk or share another traffic.

As your firewall is doing the routing to your networking, the step to add arp entry in the router isn't necessary.

The microsoft cluster that i tested worked in the "multicast" mode. I haven't tested the other options available.

Richard

Second that. mac addresses have to be statically added on NX5K switches when using MS-NLB. If the SVI for the VLAN is on the 5K, the 5K will complete arp for the servers.

When we added the static MAC to the vpc peer-link on one of the switches, the NLB would work.  If we added the static MAC to both sides of the vpc peer-link, the traffic would fail.

Has anyone else seen this behavior?  Can anyone explain this behavior?

Thanks.

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: