UCS mac address problem

[danilo-dicesare]

Hi all,

simple question....

may you know if, with release 1.4.1 there is some mac address learning bug?

i've go a blades with a mix of palo and other cards, some vnic are pinned to uplink with trunk and allowed vlan 501 on both side. On the UCS blades i've got ESX and ESXi. Service console is tagged to VLAN 501 (according to trunk configuration on UCS and uplink switch). Some blade had got these interface pinned on UCS6100 A and other one on B (with failover). The fact is that if i ping from outside UCS to inside UCS (ESX) no mac learning is performed on all infrastructure.

I'm sure problem is on UCS 'cause if i clear mac table on UCS nothing works......and all work if i begin to ping from UCS. I've also tried to put on uplink switch a static mac (ESX service console) but was not useful.

i hope is not a 1.4.1 bug......

tnx and tnx for help

dan

[tohoken]

Dan,

We are experiencing the same issue but we are running 1.3.1p so it isn't a 1.4.1 bug.  Please see my post to see if this is what you are experiencing.

https://supportforums.cisco.com/message/3263463#3263463

Ken

[danilo-dicesare]

HI,

yes i think is the same.

But is neither VLAN tag problem or arp issue i think mac address stuff. For some reason dynamic mac doesn't not respond or are not forwarded to OS installed on some UCS blade or some stuff like that....

i hope i'm wrong!

tnx

dan

[Robert Burns · Accepted solution]

Are any of you using Uplink Pin Groups or have a disjointed Layer 2 network by chance?

Issues like these are ususally related to Pin Groups/Disjoint L2 using the wrong link as the DR port for incoming traffic.  When traffic is generated from within UCS outbound, the MAC table gets updated on the upstream switch which allows traffic to flow, until it ages out.

If either of these are the case I suggest you have a look at Brad's videos, name video #6:

http://bradhedlund.com/2010/06/22/cisco-ucs-networking-best-practices/

Regards,

Robert

[danilo-dicesare]

was the case......L2 disjoint domain, but not misconfiguration

i think that could be a great thing to patch this kind of behaviour.

tnx

dan

[Robert Burns]

I believe we "might" be looking at a future enhancement to give users the ability to manually select the DR (Designated Receiver) port.  Beyond this if you require a disjoint L2 topology you can either use Switch mode, or connect the FI's to a regular L2 swtich upstream, and then uplink your separate networks from there.

Regards,

Robert

[danilo-dicesare]

hi Roberts,

tnx for answers......

Of couse for real multi customer implementation we need another layer of virtualization not just pinning (duplicate vlan id and same subnet ).

how can we implement a virtual DC with UCS? With N7K is quite clear, i can vDC chassis and duplicate all items if i've got oerlapping....

tnx and have nice day

dan

[danilo-dicesare]

another question...how can i eliminate VLAN on a uplink port?

Why do i need all vlan trunked on all ports?

tnx    

[cpaggen]

You can't manually clear VLANs from trunks today. The why is because end-host-mode is designed to be very much plug-and-play and implies you're connecting UCS to one common L2 domain. A future release should introduce the ability to prune VLANs manually.

[danilo-dicesare]

Hi Christophe,

ok for EHM but i think is not possible either in switch mode...or am i wrong?

tnx

dan

[cpaggen]

No, you're right. It isn't possible in either mode today.

[danilo-dicesare]

thx all for help.

dan