Look for Provider Backbone Bridges and Provider Backbone Transport.
Provider Backbone Bridges (aka IEEE 802.1ah, Mac-in-Mac, or MinM) overcome the service scaling limitations present into today's native Ethernet networks - which are primarily based on QinQ implementations (aka IEEE 802.1ad or Stacked VLANs). QinQ only allows for 4094 service instances per metro due to the number of bytes of the VLAN tag in the Ethernet header - not nearly enough to accommodate the anticipated subscriber base for IPTV, wireless, and other next-generation services.
Provider Backbone Bridges allow for orders of magnitude greater scalability (up to millions of service instances per metro) by physically providing more addressing space in the Ethernet frame. Additionally, it reduces the amount of state the network is required to maintain and enhances security by clearly separating the customer and provider addressing space.
Do you remember what was the Qtags for these levels?
Could we refer to EVP-LAN? We could have L2 (C-VLAN or S-VLAN) over MPLS.
Today, it's S-VLAN. Before, it was C-VLAN (the classical VLAN).
In fact, using C-VLAN over MPLS is just for the separation of the multiplexing channels in an MPLS connection to join a distinct egress PE. Each channel is associated to a VSI table in the ingress PE.
To have 4096 channels in such a connection is already important (STM-4 if each channel is 100 Mbit/s). And, if we need others channels, we can create another MPLS connection and multiplex all that in a server layer.
You can chain Ethernet switches together and get as many levels as you want, really. Think of it as an extension of QinQ.
We use 2 levels of VLAN tagging normally. In specific situations we also use 3 levels - but we try to avoid it.
There is no magic here. It's a question of scaling your network versus how much info you lose by having the switches only look at the outer VLAN tag (even when you know there are one or more inner VLAN tags).
Or, to continue with the OP's quesition ... since we have VLAN tags there is no need for an encapsulation ethertype.
Just use ethertype X'8100', add two more bytes of zeros, and you have a nicely encapsulated ethernet packet, fully conforming (even though it can be more than 1500 bytes).
Now, if you want to encapsulate including source and destination MAC address, that is another question. If I thought of a reason why you would want to do that, then maybe I would try to find an answer. Note, though, that in that case you are way over the maximum length.
Well, standard VLAN tagging uses ethertype X'8100' so it isn't right to say that there is no ethertype.
I just noticed Cisco's ISL, though, which really does encaplsulate the whole frame from destination address through FCS. It adds a
26 byte header and an additional FCS trailer for 30 more bytes.
There has always been the question about the length limit and normal VLAN (four byte) tags. Well, wikipedia seems to suggest up to three levels of VLAN, so 12 extra bytes.
Cisco seems to call ISL tagged frames jumbo, such that they might not be supported going through other devices. I don't see any description of the ISL header, but suspect that the first two bytes are a registered ethertype to avoid confusion.
outer tag was normal 802.1Q, then some Nortel numbers i dont remember
- probably related to what they pushed for provider Ethernet.
We use 2 layers of tags as do many other providers on Ethernet WAN delivery pipes.
1 tag gets stripped by an Ethernet NTE type device to flag traffic to a specific LAN port or management (the device is an Adva FSP 150 at work) - remaining tags are for VPW type point to multipoint, or just passed transparently depending on service.