Why is this subnet config required by my vlan "snake topology"?

Question

I must validate the bandwidth on all 8 ports of a switch. We intend to do this with a "snake" topology as described in this article: http://thenetworksherpa.com/test-throughput-alchemy-using-a-snake-topology/

Good news: it works. Bad news: as I understand VLANs and subnets, the configuration required to make it work should not work. I come to you all to clarify my misunderstanding.

This is a description of my setup:

vlan_12 is port_1 & port_2 and uses addresses on subnet 192.168.0.x
vlan_34 is port_3 & port_4 and uses addresses on subnet 192.168.1.x
vlan_56 is port_5 & port_6 and uses addresses on subnet 192.168.2.x
vlan_78 is port_7 & port_8 and uses addresses on subnet 192.168.3.x

All vlans tag at ingress and remove at egress. I connect the following port pairs with Ethernet cables: {port_2, port3}, {port_4, port_5}, {port6, port_7}. The combination of vlan and cables creates a single-file gauntlet for all network traffic. Traffic begins with device assigned 192.168.0.1 and connected to port_1. I will call this device ADAM.

My question is regarding the configuration of the device we connect to the other end (port_8). I will call this device ZOE. I expected ZOE would need to be assigned an address corresponding to the vlan_78's subnet. For example, 192.168.3.4. ADAM complains he can not resolve the address when he tries to ping ZOE while she is using 192.168.3.4. Stranger still, ADAM is able to ping ZOE when she is assigned 192.168.0.4. Bear in mind, I did not change the vlan_78 subnet and ZOE is still connected to port_8. I only chance the IP so it does not jive with the vlan_78 subnet. I verified all the ports are being exercised, and traffic stops when I disconnect any of the cables linking the subnets. Both are evidence the traffic is taking the path I expect. There is no evidence of shortcuts.

My questions are:

1. Must ZOE match ADAM's subnet?
2. Shouldn't ZOE required to match the subnet of the vlan she is connected to?

Please let me know if you are confident my understanding is correct and the system should not work the way I am describing. I am building a new switch from a reference design. If the switch is not behaving in ways you confirm it should, that is evidence of a bug in the reference design I need to report or investigate.

My thanks in advance. You folks are the best.

Answer 1

The point of that port/VLAN chaining rig is that you want to force the switch to receive and send in port pairs. Normally, a switch would take the shortcut to the (final) destination port which you don't want - you prevent that by placing ports in different VLANs.

By using VLAN port pairs, you define single ingress-egress port pairs - each frame that enters one port must go out the other since there's no other port. The loop cables help "convert" the (untagged) frames to the next VLAN in between pairs. (I've used the same setup once to demonstrate non-blocking forwarding on a 48-port switch.)

However, since all ports essentially form a single broadcast domain = L2 segment, you should use IP addresses from the same subnet on the edges (=for ADAM & ZOE). Configuring IP addresses on the switch (apart from management) is not required.

Note that this setup only works for L2 forwarding (bridging/switching). If you configure VLAN addresses on an L3 routing switch, the switch will simply route across VLAN12 and VLAN78 without using the ports in between. Also, you'll need to disable STP and loop detection on all ports (or use an elaborate MSTP setup).

Also note that if a vendor specifies a switch as non-blocking this kind of setup may be fun to play with but you can take full wire speed for granted.