Ignore the existence of the vlan's for a moment - assume you have three separate devices
switch_VX , switch_VY and switch_VZ with the hosts X,Y,Z connected to them.
Now assume that your router has a distinct ethernet cable to each switch.
On your router you are going to have three different ethernet ports.
eth_VX , eth_VY, eth_VZ
In this setup it should be obvious how it works .
- Each switch has it's own Layer 3 protocol address subnet for any protocols you are using (IPv4, IPv6, AppleTalk, IPX etc)
- The router needs a configuration on each interface that has an address from the same address range that the switch is using.
- Then the router gets to the host by looking at it's protocol address , looks for the interface that matches, and then uses the right protocol->mac translation mechanism to talk to the end host.
- For IPv4, that means the router looks in its routing table, finds eth_??, and then goes looking in the ARP table for the MAC address of the host it's looking for.
- Each host is configured with an address from the subnet that's on the switch it's using, and each host is configured to use the address of the router as its default gateway.
Logically that's exactly how it works - always.
However, confining people to use separate sets of switches for each subnet is not efficient. Requiring a separate ethernet cable, and port on the router for each switch isn't efficient. It gets even more costly when you want to do a proper service and add redundant cables and routers etc..
So the manufacturers changed the physical topology a bit, and moved some of the physical stuff to software configured instead. However the devices are still doing exactly the same job.
So instead of separate switches you have separate vlans on the device (or set of devices).
In the explanation above replace switch_VX with switch_vlan_X. The VLAN configuration on the switch creates effectively a completely seperate switch. It runs its own MAC Address Table, it's got its own copy of spanning-tree running. Internally inside the switch it has to record the VLANID inside each ethernet frame - so that it makes sure that it never gets sent out the wrong ports. The switch adds the VLANID when it receives a frame, and strips it off before it sends it out. So the end hosts have no idea that it's happening. It's all hidden.
So that removes the multiple switches, and we can configure our vlans on one switch.
But we still have multiple cables to our router. So let's fix that by configuring the switch engine to treat the port connected to the router as special. Instead of stripping off the VLANID for all frames - let's instead send the frames up to the router with the VLANID still on them. We'll need to agree on a common format for the frames, so the routers know where to look. 802.1Q is the industry standard, but there are some other options out there. Most vendors call the port a ''trunk'' port when it is configured to leave the VLANID in place.
Now the router is getting a stream of frames on a single interface but they have VLAN identifiers in there that need to be removed. Let's get the router to do that in software.
So in the description above, instead of different ethernet interfaces ; we'll have a software interface that understands VLANs. Replace all mentions of eth_VX with eth_vlan_X .
Now the router knows when it gets a frame that is part of VLAN X, that it is associated with the interface eth_vlan_x , and it can remove the VLANID and process it appropriately.
If the router wants to send a frame out the interface eth_vlan_x, it knows that it needs to insert the VLANID X into every fraame.
So we started with a logical setup, and changed the physical layout to be more flexible and more efficient. However, logically it is absolutely no different to the setup that uses independent separate devices.
