ACI FI VLAN

[mgual]

Hello,

We have an aci fabric in multipod with UCS and VMM integration.

2 leaf are connected to 2 fabric interconnect via Vpc.

 What will happen when we reach the vlan limit(3960) on both leaf and fabric interconnect ?

Do we need to buy other Fabric interconnect?

[Robert Burns]

Without knowing your design, pushing 3900+ VLANs down to the FIs is pretty huge.  Do you absolutely need every single ACI EPG available on UCSM?

In a network-centric deployment (1 EPG per 1 VLAN), you're looking at a Leaf Max of 3960 EPGs - even though UCSM can support 4K VLANs with ACI (which is a combination of regular & private VLANs).  So adding another pair of FI's to the same pair of Leafs isn't going to help you here.  Your best option would be to only push EPGs/VLANs to the UCS domain that are actually required.  Don't just deploy VLANs everywhere just because.  You'll easily  blow out your Leaf scale. 

Beyond this, the only way to scale beyond this would be to separate your compute capacity across another pair of Leaf nodes. 

Robert

[mgual]

Hello,

In fact for each EPG, we have a virtual router behind UCSM acting as the default gateway, so each EPG need to be present on UCSM

So I dont have the choice

Regards

[Robert Burns]

In that case you're limited to what I stated above. 

Robert

[jgomezve]

Hello 

You should have a look at at the VMM integration with AVE. This integration natively uses VXLAN packets between the Fabric and the Hypervisor virtual switch. Thus the only VLAN you will need to enable in your UCS infrastructure, namely the FI, is the ACI Infra VLAN

[Robert Burns]

Where the AVE would make the programming of intermediate devices more simplified, it wouldn't address his primary scale concern.  Still constrained by the leaf resource limits.

Robert