MPLS or SD-WAN: linking your sites to your data center
The question is wrongly framed nine times out of ten. MPLS and SD-WAN are not two competing products to be picked from a comparison grid: one is a transport service sold by a carrier, the other is a way of steering links you buy elsewhere. When your applications live in a data center cabinet, the real question is what happens at the far end. This guide separates the two subjects.
Estimate my cost →What MPLS does, in one honest sentence #
MPLS, standardised in RFC 3031, switches packets on a label rather than an address. That technical detail has a commercial consequence: the carrier controls the path end to end, so it can sell a commitment. Honoured service classes, guaranteed bandwidth, contractual restoration time. You are not buying a pipe, you are buying a promise kept by a single company, which explains both the price and the lead time.
What SD-WAN does, in an equally honest sentence #
SD-WAN builds encrypted tunnels, usually IPsec, over any access: consumer fibre, dedicated link, 4G or 5G. A controller continuously measures loss, latency and jitter on each link and steers each application to the best one available. That is powerful, and here is the crucial nuance: SD-WAN MEASURES and SWITCHES, it does not RESERVE anything. On a shared access, nobody owes you bandwidth.
The difference comes down to one word: guarantee #
With MPLS, quality of service is reserved inside the network: a voice class stays protected even when the link is loaded. With SD-WAN over shared access it is observed: you know the link has degraded and you move to the other one, but if both are degraded there is no third. That is not a flaw in SD-WAN, it is the nature of what you bought underneath. SD-WAN over two committed dedicated links gives you both, and costs accordingly.
The trap everyone forgets: the link underneath #
Comparing MPLS and SD-WAN without looking at the underlying access makes no sense. In both cases the traffic uses the same local loop and often the same backhaul. SD-WAN over two consumer accesses delivered by the same infrastructure operator, from the same exchange, is not redundant: it is one point of failure with two invoices. Check physical paths before counting links.
The real decision point is at the data center end #
Your sites talk to servers, and those servers are in a cabinet. The network concentration point is therefore the data center, and three questions arise there. Is your carrier physically present on that site, or must you pay for transport to bring it in? Is the connection made by a cross connect to its meet-me room, and at what recurring cost? Does the site host enough carriers for you to change your mind in three years without moving your servers? A carrier-rich data center sometimes costs more per cabinet and far less over time.
How to choose, concretely #
- Real-time flows matter to you, daily telephony or video, and a thirty-minute outage has a price: buy a guarantee, MPLS or a dedicated link, at least on the sites that carry the business.
- You have many small sites, irregular needs and tolerance measured in minutes: SD-WAN over shared access will cost three to five times less, and central management alone justifies the move.
- You are unsure: the answer is almost always hybrid, guaranteed links on the few critical sites and headquarters, managed shared access everywhere else.
- In every case, demand the data center connection detail line by line. That is where the recurring fees nobody mentions in the first proposal hide.
- Ask for the lead time BEFORE the price. A guaranteed access on a poorly served site can take several months and civil works billed separately.
The questions to put to your carrier #
What is the guaranteed rate, and what is the peak rate? Does the guarantee cover only the access link or the whole path to the delivery point? What is the restoration time, over which hours, and with what penalty if missed? Does the second link follow a genuinely different physical path, and will you put that in writing? Where do you deliver my traffic inside the data center, and who pays for the cross connect? Those five questions beat every comparison table on the market.
What to remember #
MPLS buys a guarantee, SD-WAN buys flexibility and visibility. Neither creates capacity where there is none, and neither compensates for a weak connection at the data center end. Start with the part of the chain you will not be able to change easily, that is the site where your servers live and the carriers present there. The rest can be renegotiated and replaced.
FAQ #
It often replaces the way MPLS was used on secondary sites, where the real need was reliable connectivity rather than a contractual guarantee. On sites carrying critical real-time traffic, the guarantee remains a separate product that SD-WAN does not manufacture.
Transport is encrypted, usually with IPsec, which is not the case with MPLS where isolation is logical rather than cryptographic. The security difference comes mostly from the surface exposed to the internet and the quality of administration, not from the technology itself.
Not necessarily. Two commercial carriers may resell the same local loop, pass through the same exchange and share the same duct. Redundancy is verified on the physical path, not on the supplier's name.
It is far preferable. If it is not present, transport to the site must be funded, with a recurring cost and often a long lead time. The number of carriers on a site is a data center selection criterion in its own right.
Yes, and it is a common design. MPLS becomes one of the underlying links, with its guarantee, and the SD-WAN controller arbitrates between it and a cheaper internet access depending on the application.
Written on 6 September 2026.
From reading to comparing: relevant data centers
Estimate my cost → Compare data centers