Cat9500 StackwiseVirt Woes

[casanavep]

So, I have noticed issues with several Cat9500, where only ports on the active switch are utilized in LAGs spanning members. This occurs when traffic crosses an SVI boundary. Here's the kicker, Cisco says it is as expected (SR686413667). This problem exists on 16.9, 16.11, and 16.12 code. Here's the scenario, imagine two 10Gbps internet pipes coming into a site. You terminate one into each switch, so you can survive a loss of either two-switch stackwise-virt SSO members. You then LAG via one 10G link from each to your next device in the environment, maybe a firewall, load-balancer, or NAT device. If the traffic crossed a layer-3 SVI boundary between the internet ingress and the egress LAG to the next device, traffic from the standby member will be passed over the SVLs to the active switch, and only utilize physical interfaces on that switch on egress. So, even though your egress path exist of 2x10G LAG members, only one will be utilized due to the other being on a standby switch. This results in huge numbers of output drops and serious limitations in HA. As designed....

[Reza Sharifi]

This is seems like a design issue as the StackWise supposed to work like VSS where each switch forwards traffic. What happens 

Also, you may have to break the StackWise design and just use each switch as a standalone switch and /30 from each switch to the provider router and a /30 between the switches. For outbound traffic use HSRP or VRRP.

HTH

[casanavep]

We have verified that breaking the stack and using routing protocols and CEF for load-balancing works.  That said, what's the purpose of StackWiseVirtual if I cannot scale to utilize the link-aggregation strengths of a single logical forwarding plane? 

[casanavep]

I should add that we tested several switch OS versions and models.  Cisco 4500X switches also appear to have this behavior.  I am guessing there is some design flaw commonality between the two.