Half of blades were down when a new uplink port-channel was created?

[gwhuang5398]

I have a pair of 6248UP fiber interconnect (A and B) with blade servers connected. The upstream switch is Nexus7k. Right now Fabric A and B each has a uplink port-channel to the upstream Nexus switch. I'm adding a new Nexus switch to connect to the fiber interconnect. The plan is to have some VLAN's uplink to one Nexus switch and some VLAN's uplink to the second Nexus switch.

After I configured the new Nexus switch with a port-channel (only allowing some VLAN's), and created matching port-channel on Fabric A and B, new port-channels were up. However half of the blade servers were down after that. What could be the reasons for this problem? Is it because half of the blades were dynamically pinned to the new port-channel but the new port-channel on the Nexus side only allow some VLAN's?

Is there a way to configure the port-channels on fiber interconnect to only trunk through some VLAN's? so server traffic would not be sent to the port-channel where the server's VLAN is not allowed?

Thanks                    

[Walter Dey · Accepted solution]

In general, all vnics/vhba's of a fabric are down, if all the uplinks of this fabric are down.

It is not clear to me, what you try to achieve ?

If you create a global vlan (for both fabrics), this vlan's are automatically inserted in all the North bound trunk links, unless you configure a disjoint setup

(http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns944/white_paper_c11-692008.html)

Creating a new vlan on UCS requires of course also, that this vlan is created on the connecting switch, and vpc by default allows all vlans. Could it be, that the vlan was not created on the switch, and therefore was not up on the trunk ?

[gwhuang5398]

Thanks a lot. I think you answered what I was looking for. I need to configure disjoint upstream. The reason for this is I have some VM instances on the UCS in one set of vlans and some other VM's in another set of vlans. The two sets of vlans will go through different upstream security devices.

When I first configured the new port-channels to UCS, I didn't tie them to a set of vlans. So maybe half of the VM, reagardless of vlans, had traffic sent to that new port-channel. Because the new switch was fully not in the network yet. The traffic really had nowhere to go after that. So those VM's were seen as down.

Thanks again.