Microsoft NLB with UCS & Upstream Nexus 7000

[patelvc7601]

hi

I know lot of users is using NLB & My concern is if everybody can chip in with their idea we can make this question & convert to the document .

We are trying to setup the Microsoft NLB with UCS  Enviornment. I have 3 Chassis connected to 6120 which in turn connected to vPC to pair of nexus 7k.

I have configured NLB with multicast option with Sinlge NIC. Multicast Mac address is 03-BF-XX-XX-XX-XX ( where xx is mapped digit of Ipv4 of cluster Ip)

I have configured static arp  on the Nexus 7k with VIP :

LB1 IP/MAC                   VIP ( Virtual IP or Cluster IP)

10.10.10.11                 10.10.10.10

0050.5697.0001           03-BF-0A-0A-0A-0A

LB2 IP/MAC

10.10.10.12

0050.5697.0002

I have static arp on  nexus 7k with 10.10.10.10 map to 03-bf-0a-0a0a-0a

Before static arp I could not ping cluster ip from outside this subnet ( well known issue with NLB )

After creating static arp I could ping cluster ip & nlb seems to work but when I do the Etheral capture on Lb1 & lb2 ,it shows me destination mac as 0050.5697.0001 on lb1 & lb2 0050.5697.0002 . on 6120 , I do not see any mac address reference to 03-bf-... , I am puzzled how this thing working.

Sincerely,

Viral

[Mathew Lewit]

I am not sure if you are aware of this issue on UCS.

http://tools.cisco.com/Support/BugToolKit/search/getBugDetails.do?method=fetchBugDetails&bugId=CSCtg82066

But in order for multicast to work you will have to enable the querier upstream (not sure if you have done this or not).

[patelvc7601]

Hi Mathew,

thank you for the reference bug id . I have set NLB cluster mode with multicast ( there are 3 option , unicast,multicast & igmp multicast ). Igmp multicast option use standard 01:00:5e: format while multicast option use that 03-bf-xx- mac address.

As far as I know igmp snooping will control how multicast traffic forward at layer 2 ( it will snoop the conversation & flood to only port that are interested). Regardless of it, I am not using igmp multicast mode .

Weird part is I have arp set for virtual ip 10.10.10.10 to 03-bf- & UCS does not know about this mac address , it knows individual lb1 & lb2 mac address. According to

http://bradhedlund.com/2010/10/22/cisco-ucs-qa-2-end-host-mode-forwarding-behavior/

unknown mac Fabric Interconnect will drop the packet. In my case it's forwarding the traffic & when I do Ethereal Capture at the lb1 & lb2 I am seeing the mac address of lb1 & lb2 ..

Sincerely

Viral

[wuhao_xiaotong]

Have you configed the "peer-gateway" option at the vPC ? I have the experience about that when used vPC in front of the F5 NLB device.

Sent from Cisco Technical Support iPad App

[patelvc7601]

Hi Gary,

I have peer-gateway enabled for my NAS Storage & this is not a issue at all here, I have NLB working in this instance.

I could not figure out how when I have static arp sate on Nexus 7k for 10.10.10.10 to

03-bf-0a-0a-0a-0a & ping works fine but when I do ethereal Capture on LB1 or LB2 it shows LB1 & LB2 mac address .\

Thank you

[davidcheung]

As the LB1 and LB2 never use the cluster MAC to send any traffic out, all layer 2 devices wont see the cluster MAC as source MAC of ethernet frame. So you wont see cluster MAC in their MAC address table.

Cheers