Backhaul, L2TP and VLAN: what really lies behind your link
Your quote shows one line and one bit rate. In reality your traffic crosses four segments and at least two companies before reaching your cabinet. That split is not a detail: it explains why an advertised rate is not always reached, why an outage can drag on even with a responsive provider, and why two seemingly redundant links are not. This guide describes the real path, without needless jargon.
Estimate my cost →The path, segment by segment #
The first segment is the local loop: the copper pair or fibre linking your building to the neighbourhood exchange. It belongs to an infrastructure operator, almost never the one that invoices you. The second is backhaul: carrying your traffic from that exchange to a point where your commercial provider collects it. The third is handover, at that provider. The fourth brings it to your data center cabinet.
L2TP: how a provider sells you a link on a network it does not own #
On copper access, and on some fibre offers, the historic mechanism is L2TP, described in RFC 2661. Your session, opened over PPPoE from your router, is picked up by a concentrator at the infrastructure operator, then tunnelled to an access server at your commercial provider. That provider authenticates you and assigns your IP address. In other words, it can sell access anywhere without owning a metre of fibre, which is commercially very effective and lengthens the technical chain.
VLAN: the other approach, more common on fibre #
On fibre access delivered over Ethernet, customers are separated by VLAN, as defined by IEEE 802.1Q. Depending on the offer, double tagging is used: one tag identifies the backhaul area, a second identifies the subscriber. There is no session to open, the link is simply a virtual cable between customer and handover point. It is easier to operate, and the achievable rate then depends entirely on backhaul sizing.
Why your bandwidth is not always there #
On a shared offer, the local loop is rarely the bottleneck. It sits further along, on the backhaul link shared between many subscribers, or on your provider's handover port. That is why an excellent speed test at three in the morning and a poor one at eight in the evening reports no fault: it reports sizing. A dedicated access does not behave that way, and that is precisely what the higher price buys.
The symptom that gives the chain away: MTU #
Encapsulation costs bytes. A PPPoE session typically brings the usable packet size down to 1492 instead of 1500, and further tunnelling eats more margin. When nothing adjusts the size announced by the endpoints, the symptom is distinctive: basic network tests pass, some sites load, others hang half-rendered, and a file transfer stalls. If you see that picture after switching provider, this is almost always the culprit.
What the split changes for your restoration commitment #
- The commitment you sign covers only the scope of whoever sells it. A local loop fault is handled by a third party, on its own timescales.
- The number of companies in the chain predicts real repair time far better than the figure printed in the contract.
- Always ask which wholesale offer is used, activated or unbundled. The answer tells you who physically turns up when the line is down.
- Two links invoiced by two different providers may share the same loop and the same backhaul. Require in writing that the physical paths differ.
- On a critical site, the only real backup is an access whose local loop is of a different nature or from a different infrastructure operator, ideally entering through another side of the building.
And at the data center end #
The last segment is usually the easiest, because in a data center the carriers are already there. Connection is made by a cross connect to their space, at a recurring fee, and lead time is counted in days rather than weeks. That is the exact opposite of the office situation, where the last mile is expensive and takes months. It is one of the reasons concentrating servers in a well-served site simplifies the rest of the network.
What to remember #
A carrier link is not a wire, it is a subcontracting chain. Knowing where your traffic changes hands tells you who to call, what your commitment really covers, and whether your redundancy is real. Three questions get you the essentials: which wholesale offer, which handover point, and which physical path for the second link. A serious provider answers all three.
FAQ #
It is the transport of your traffic between your neighbourhood exchange and the point where your commercial provider collects it. It is the segment you never see, and where most bandwidth limits on a shared offer are found.
Because the local loop belongs to an infrastructure operator. Your provider raises a ticket with it and depends on its timescales. That is why a restoration commitment must state what it covers.
With unbundling, your provider installs its own equipment at the exchange and controls more of the chain. With an activated offer it buys a ready-made service and simply collects the traffic at the handover point, with fewer technical levers.
The tunnel itself costs little. What is felt is backhaul sizing, access server load at peak hours, and above all packet size problems when nothing adjusts the value announced by the endpoints.
Ask in writing for the infrastructure operator of each access, the exchange used, and the entry point into the building. If any one of the three answers matches, the redundancy is partial.
Written on 6 September 2026.
From reading to comparing: relevant data centers
Estimate my cost → Compare data centers