SD-WAN or MPLS? Choosing a WAN for Multi-Branch Companies

SD-WAN or MPLS? We compare them on SLAs, application-based path selection, line diversity and the routing and MTU problems that can come up during migration.
Technical summary
Technology
SD-WAN, MPLS L3VPN, IPsec, BGP, ZTP
Topic
Choosing a multi-site WAN architecture
Level
Intermediate
Prerequisites
  • Basic IP routing knowledge
  • Branch and application inventory
In this article8 sections

If branch traffic is pulled to headquarters and goes out to Microsoft 365 from there, users say “the internet is slow,” and they’re right. The problem is often not line speed but a WAN architecture that doesn’t account for where applications now run. The choice between MPLS L3VPN, IPsec over the internet and SD-WAN isn’t a line order; it’s an architectural decision made on where applications are, what infrastructure branches can get and who will operate the network. Below we compare the three models on SLAs, path selection, line diversity, management and cost items, and give the problems you may encounter during migration and a decision table by branch profile.

Classic WAN models and their limits

There are three classic models for the WAN in an enterprise network infrastructure. The first is the L3VPN service the carrier provides over its MPLS backbone: branches connect to the carrier’s PE routers, and routing is shared with the carrier. The second is site-to-site IPsec tunnels built over the internet line at each location. The third is connecting locations at layer 2 with the carrier’s Metro Ethernet service; the VLAN, MTU and bandwidth profile details at the handoff point are explained in the Metro Ethernet E-Line and E-LAN article.

The common weakness of all three is that they’re unaware of applications. A router doesn’t know whether a packet is ERP, voice or the nightly backup; it looks at the destination network and picks a single path. That was enough while applications ran in the data center at headquarters. As the weight of SaaS and the cloud grows, carrying branch traffic to headquarters first and then out to the internet (backhaul) means double latency and unnecessary load on the headquarters line. The headquarters line fills up.

MPLS L3VPN: carrier SLAs and constraints

MPLS L3VPN (RFC 4364) keeps each customer in a separate VRF in the carrier backbone. The CE router on the customer side shares routes with the PE via static routing or eBGP; traffic between branches stays in the carrier network. Its strengths are:

  • End-to-end SLA: Because traffic stays in a network controlled by a single carrier, contractual commitments can be given for latency, jitter, packet loss and availability.
  • Carrier QoS: DSCP-marked traffic can be mapped to priority classes in the carrier backbone too. The number of classes and the mapping vary by carrier; get them in writing before the contract.
  • Predictable behavior: It isn’t affected by routing changes on the internet.

The constraints are clear too. MPLS separates traffic with labels but doesn’t encrypt it; if confidentiality is required, you need to add encryption on top. Commissioning a new location depends on the carrier’s infrastructure access at that address and can take weeks. The unit cost of bandwidth is usually higher than internet access. Dependence on a single carrier narrows both your price negotiation and your options during a fault. A central internet exit also needs to be designed for cloud applications.

The most common misconception we see in the field about MPLS is the comfort of “it’s the carrier’s network, the carrier will handle it.” If the QoS marking on the CE side is wrong, the carrier’s priority classes are of no use. You do the marking.

SD-WAN: overlay, central policy and application routing

SD-WAN builds an encrypted overlay on top of the physical lines and manages this overlay with policy from a central control/management plane. The components are:

  • Edge devices: Devices at each location that connect to more than one underlay line (fiber internet, MPLS, 4G/5G) and build IPsec tunnels between them.
  • Control and management plane: The component that centrally distributes the topology, tunnel keys, routing and steering policies; depending on the vendor, it runs in the cloud or on premises.
  • Path measurement: Edge devices continuously measure latency, jitter and packet loss on each tunnel with active probes.
  • Application recognition: Which application traffic belongs to is determined from the first packets, DNS/SNI information or a signature database.

When these come together, policies can be written such as “move voice traffic to the other path from a path whose one-way latency exceeds 150 ms (the upper limit ITU-T G.114 gives for interactive voice) or whose loss exceeds the set threshold.” SaaS traffic exits to the internet directly from the branch (local breakout), and internal applications go to headquarters through the tunnel. At a new branch, the device connects to the management plane with zero touch (ZTP) and receives its template.

Don’t buy SD-WAN without knowing its limits. SD-WAN doesn’t improve the line itself: if two lines pass through the same building in the same duct or connect to the same carrier exchange, the common failure risk stays exactly the same. On the path across the internet there’s no SLA between carriers; SD-WAN only chooses the best of the paths it has. The management plane is vendor-specific; devices from different vendors generally can’t build an overlay with each other. The MEF 70 standard defines the characteristics of the SD-WAN service offered by carriers; it doesn’t provide interoperability between devices. Don’t plan a multi-vendor overlay.

An SD-WAN mistake you may encounter is leaving SLA thresholds at the product’s default values. The default threshold doesn’t know your application. Set thresholds according to the values you measure with real traffic at the pilot branch.

Hybrid WAN: MPLS and internet together

In practice, most companies don’t make a binary choice of either MPLS or SD-WAN. In a hybrid WAN, the branch has both an MPLS and an internet line, and the SD-WAN edge device uses both as underlay. Because the overlay tunnel is built over MPLS too, policy and encryption are the same on both paths. A typical split:

  • Voice, central video conferencing components and latency-sensitive internal applications: primary path MPLS, backup path internet.
  • SaaS and general web traffic: the direct internet line; if it degrades, the headquarters exit through the tunnel.
  • Backup and file synchronization: the lowest-cost path, bandwidth-limited during business hours.

This model lets you shift growing traffic to the internet line instead of upgrading MPLS capacity. Over time, MPLS is kept only at critical locations or removed entirely. The decision is made with measurement data.

Comparison table

Criterion MPLS L3VPN Internet + IPsec SD-WAN
End-to-end SLA Contractual None Depends on the underlay
Encryption Not by default Yes Yes
Application-based path selection None None Yes
Line diversity Single carrier Limited Multiple carriers and cellular
Cloud access Through headquarters Local or headquarters Local via policy
New location Depends on the operator Manual configuration Template and ZTP
Management model Carrier + customer Per device Central policy
Cost items Line and port fees Line + device Line + device + license

Comparing cost only by the monthly line fee is misleading. The device refresh cycle, subscription-based licensing, the labor of the team operating the network and the business impact of outage durations should go into the same calculation. SD-WAN licenses are mostly priced by bandwidth tier and feature set; at the bid stage, ask how the cost will change when a tier upgrade is needed three years later.

Migration phases and risks

Moving from MPLS to SD-WAN or a hybrid model isn’t done overnight. The order below keeps outage risk under control:

  1. Inventory and traffic analysis: The line type, capacity, contract end date and application-based traffic distribution (with NetFlow/IPFIX data) of each location are collected. Start with measurement, not estimates.
  2. Underlay preparation: Second lines are obtained, where possible, with a different infrastructure and a different building entrance. The dual internet line and failover article covers line diversity and failover behavior in detail.
  3. Headquarters and data center edges: The hub side is built first; route exchange between MPLS and the SD-WAN overlay (usually with BGP or OSPF) is designed here.
  4. Pilot branches: Two or three branches from different profiles are chosen; application policies and SLA thresholds are validated here.
  5. Rollout in waves: Measurements are compared after each wave; MPLS contracts are closed only after the new structure runs stably.

The problems that come up again and again in the field during migration are well known. Routing loops and asymmetric paths that form while redistributing routes in both directions between MPLS and the overlay; a stateful firewall drops the session when it sees outbound and return traffic on different paths. The MTU overflows because of the overlay tunnel header; small packets pass, large file transfers stall. Application recognition rules work less accurately than expected on encrypted traffic. And the default SLA thresholds mentioned above. The asymmetric path is the first place to look. How to measure line quality is explained in the WAN SLA measurement article.

Decision table by branch profile

Branch profile Recommended model Reason
Small office, cloud-heavy SD-WAN, two internet lines or internet + 4G/5G Traffic goes to the cloud, not headquarters
Medium branch, ERP and IP telephony SD-WAN, internet from two different infrastructures Path selection protects voice quality
Production facility dependent on headquarters systems Hybrid: MPLS or Metro Ethernet + internet SLA-backed primary path, independent backup
Data center / headquarters Multiple carriers, SD-WAN hub The point where all branches converge
Temporary or field location Cellular + SD-WAN or IPsec Fast setup, no fixed line

Before deciding, answer these four questions: How much of the applications are in the cloud? Is there a second infrastructure provider at the branch? Is a contractual SLA mandatory for the business? Who will operate the network, and with which tools? Moving to product selection before these questions are clear means letting the product decide the architecture. How the branch’s own internal network will be standardized is the subject of the standard branch network design article.

For the inventory of existing lines and carrier options, see the Metro Ethernet and internet line management page; for the architecture decision and migration plan, the enterprise network consulting page.

Frequently asked questions

Can SD-WAN completely replace MPLS?

If most applications are in the cloud and branches can get internet lines from two different infrastructures, in most cases it can. If there are applications very sensitive to latency and jitter, a need for a contractual end-to-end SLA, or locations where line diversity can’t be achieved, a hybrid model is a safer intermediate step.

Is an MPLS line encrypted?

No. MPLS L3VPN separates traffic with labels in the carrier backbone but doesn’t encrypt it. The separation is logical; if data confidentiality is required, use IPsec or a similar encryption layer over MPLS too.

Does SD-WAN require two internet lines at every branch?

Technically it works with a single line too, but then there’s no alternative path for application-based path selection and the gain is largely limited to central management. At critical branches, get a second line from a different infrastructure or a cellular backup line.

In a hybrid WAN, should the MPLS line be kept on standby?

Usually not. Both paths are kept active; voice and critical business applications preferably go to MPLS, and cloud and general internet traffic to the internet line. When one path degrades, policy moves traffic to the other. A line waiting on standby is both idle capacity and an untested path.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Network

We build switching, routing, cabling and wireless networks with VLAN segmentation and redundant uplinks, and provide regular maintenance and fault support.

Knowledge Center

Other articles