Currently we have CSS deployed with two arms to the same L3 distribution switch. One arm, lets call it the external arm, is bridging a single vlan, with multiple networks configured under the Circuit VLAN. One of those networks is used for static routing client traffic (request and reply) between the css and switch. The other networks are different VIP networks which only have presence on the CSS.

The 2nd arm, lets call it internal arm, is another bridge also with one vlan defined. This arm is meant for traffic to load-balanced servers from the CSS. The servers live on multiple VLANs connected to the switch, all reachable via static routing from the CSS to the switch over this second arm. The server default gateways are the L3 switch itself.

We mostly PBR the server traffic on the L3 switch, back to the CSS for TCP consistency. Does this model fall into either the one-arm or inline models ? Or is it another model that we invented. Thanks.