ACI VMM integrate with HP-Blade Chassis: uplinks

[rubberheart]

Hi Team , 

we have pushed ACI-DVS into VCenter successfully. we have 8X10G Leaf ports for the whole chassis uplink. so in the ACI-DVS switch shows there 8 uplinks in VM Vcenter side.  

here is the question from VM-Admin:

each blade will only have 2 VMNIC (0 and 1) , on normal VM-DVS , only have 2 Uplinks and they just associate the VMNICs with Uplinks. 

now comes to ACI-DVS , how to associate 2 VMNICs to 8 Uplinks?

I guess the chassis will have 16 blades , if all blade VMNIC0/1 are only associated with uplink0/1 , what about other uplinks(3-7)?

kind of confusing now , please help 

much appreciated !

regards

Daniel

[gmonroy · Accepted solution]

Daniel,

The 8 uplinks presented to vCenter within the DVS do not have a direct mapping to ACI leaves assuming that is the association you are making. For example, in my lab I have 4 leaves but still see the 8 uplinks which looks to be what the APIC pushes to vCenter by default:

 and you can see that I have 5 hosts (2 nics per host) associated to uplink 1 and 2 respectively with no issue. 

Looking through the DVS settings, we see the following in regards to the number of uplinks ports:

Where it defines that number as the max numbers of physical adapters per host. So in your setup, this should have no real bearing on communication or allocation of host uplinks to DVS unless you end up having more than 8 adapters per host that you want to exist on this single DVS.

Are you seeing some specific issue with communication in your setup?

-Gabriel

[rubberheart]

Thanks Gabriel , we do have issue with this HP settings

those VLANs are comes and goes , flapping , which cause the traffic up and down .

from ESXi host side , we normally should see the cdp neighbor is the LF interface , but for some reason it learn its neighbor with other ESXi-host (there are 16 ESXi-host blade in that Chassis)

we have opened a case with HP about this

in the meantime , do you have any idea why ? 

much appreciate!

Daniel

[gmonroy]

Daniel,

In your first post you mentioned that the Servers in question are part of some blade-chassis type of topology? If that is the case, then my first thought is that the neighbor of the Blade server should be the intermediate switch (in the case of a UCS-B blade chassis environment, the Blade server neighbor would be the Fabric Interconnect).

If the above is true, then it is definitely strange if the Blade server neighbors are instead reporting as being another blade server.

As for your first point, depending on the Resolution Immediacy of the VMM domain associated to the EPG, we can potentially be relying on this neighbor information to be correct since that is what ultimately determines when the VLAN should be programmed on the leaf interface.

For more information on this behavior, please read the following doc:

ACI Fundamentals Guide - Resolution and Dployment Immediacy

-Gabriel