TECHNICAL GUIDE · GLOBAL ARCHITECTURE

Enterprise SD-WAN with BGP Anycast and Multi-Homed Edge Routing

Automating sub-second failover between fiber carriers, 5G cellular, and low-earth orbit satellite links at branch locations.

Updated october 2026
8 min read
Altovar Technical Editorial Team

1. Why branch connectivity needs more than one link

A branch that depends on a single internet connection stops when that connection fails. Most outages, however, are not clean failures. More often a link degrades: packets are lost, latency jumps, or a route works for some destinations and not for others. A single line can hide that risk even when its contract looks reliable on paper. Because the failure is partial, it can last longer than a full outage and frustrate users more.

Multi-homing means connecting a site to more than one provider or access type, such as fiber, cellular and satellite. The goal is continuity for critical applications, not identical speed on every link. Decide which applications need which path. A cellular link with usage limits or a satellite link with higher latency may be a good backup for some traffic and a poor primary choice for voice or point-of-sale transactions.

2. How SD-WAN and BGP divide the work

Software-defined wide area networking (SD-WAN) is a software layer that measures the quality of each path and steers application traffic over the best available one. It sits above the physical links and applies policies, for example sending voice over the most stable path and bulk file transfers over the cheapest one. A central controller usually distributes these policies to the branch devices, so changes do not need to be made by hand at each site.

Border Gateway Protocol (BGP) is the routing protocol that exchanges reachability information between networks. When a company announces its own address space to more than one provider, BGP influences which path the wider internet uses to reach it. Anycast means that the same address is announced from several locations so that traffic reaches the nearest or best one. Anycast is mainly used for services delivered from several sites, not for ordinary branch traffic. Keeping these roles separate avoids designs where two systems fight over the same routes.

3. Comparing redundancy designs

Each design trades simplicity against control. A manual backup line is cheap but depends on someone noticing the outage and acting. Active-standby links switch automatically when a link goes down, yet they may ignore degradation. Quality-based SD-WAN reacts to measured loss and latency, which is closer to what users experience. Multi-homed BGP offers the most control over routing, and it requires the most skill and discipline.

The table gives a qualitative comparison. Most branches with real dependence on connectivity benefit from quality-based steering, while organizations with their own address space and experienced network staff may justify BGP. Choose the design that your team can operate and test, because an elegant design that nobody can troubleshoot fails in the worst possible moment. Write down the reasons for the choice.

DesignFailover behaviorOperational effortTypical use
Single link with a manual backupManual switch after an outageLowVery small sites with low dependence
Active-standby dual linksAutomatic on link down, degradation may go unnoticedMediumSites with a reliable second provider
SD-WAN with quality-based steeringMoves traffic based on measured loss and latencyMedium to highBranches with mixed application needs
Multi-homed BGP with own address spaceRouting changes after the network reconvergesHighOrganizations with skilled network staff

4. Common mistakes in branch routing

The most common mistake is buying two links that share the same physical route, such as the same duct, the same street cabinet or the same upstream provider. A single cut or a single carrier incident then takes both down, and the business assumes it was protected. Ask where each circuit physically enters the building and whether the carriers depend on the same infrastructure. Two contracts are not automatically two paths.

Another mistake is measuring only whether a link is up or down. Failover policies should use loss, latency and jitter for each application class, because a link that answers pings may still ruin a video call. Also test asymmetric routing. Return traffic may take a different path than outgoing traffic, which can break stateful firewalls unless they are designed for it. Plan how sessions are kept alive during a switch and confirm this in a real test.

5. Checklist for a branch redundancy design

A redundancy design is only as good as its tests. The checklist below starts with the business view, which applications matter and how much interruption they can tolerate, and then moves to the physical and logical details. Each item should be confirmed in writing by the carrier or the provider, not assumed from a sales document. Keep the signed answers with the design record.

Repeat the checklist whenever a carrier changes, a new application is added or the branch moves to a different building. Physical routes and contracts change more often than most teams expect, and a redundancy design that was correct two years ago may now depend on a shared path. Review the carrier contracts at the same time, because they are the place where route details should be written down.

  1. List the applications at the branch and rank them by business impact.
  2. Confirm which access types are available, and where each circuit enters the building.
  3. Check that the carriers do not share the same upstream path where that can be avoided.
  4. Define quality thresholds for each application class, not only for link state.
  5. Set the failback policy: whether traffic returns automatically or only after approval.
  6. Verify firewall and address translation behavior under asymmetric or changed paths.
  7. Test a real cut by disconnecting each link during a planned window, and time the recovery.

6. What to ask a supplier or internal team

Ask for the physical routing of each circuit, not only the carrier name. Ask whether the cellular and satellite links have usage limits, latency characteristics or traffic restrictions that rule out certain applications. Ask who manages the BGP sessions, if any, and how address announcements are approved and documented. A short answer to these questions often reveals that a design depends on one shared path.

Ask how failover is measured in the proposal. If the answer promises very fast recovery, ask which applications were tested, under which conditions, and what the impact on active sessions was. Ask how the provider monitors the branch, who receives alerts and what the escalation path is when a link degrades but stays up. Monitoring that only reports outages misses the failures users feel most.

7. How to start with one pilot branch

Pick one branch with a clear dependency on connectivity and an existing second link or an easy way to add one. Document its current applications, their criticality and the route each one uses today. This baseline shows where traffic steering matters most and whether the second link is really independent of the first. Note any shared equipment, such as a common router or power supply.

Run a controlled test: degrade the primary link first, then cut it, and observe what each application experiences. Repeat the test after each change. The first concrete step is to ask the current provider for the physical path of the existing circuit and to confirm in writing whether any second line uses the same route. Keep the results with the design record.

Frequently Asked Questions

Is sub-second failover realistic for every application?

No, not for every application. Some sessions survive a path switch when the application retries automatically, while others drop. Failover time depends on how quickly the path is detected, on the routing protocol and on the application itself. Ask for test evidence using your own applications, not a generic figure.

Can a 5G or satellite link be the primary connection?

It can be for some sites, if the traffic profile fits. Cellular and satellite links often have variable quality or capacity limits. Many organizations keep them as backup links or as a path for selected traffic. Test under realistic load and at the real site before relying on them for critical work.

Do we need our own BGP address space?

Usually not. Most branches use addressing supplied by their providers or an SD-WAN overlay. Running your own BGP setup makes sense when you have several providers, your own address space and staff able to operate it safely. Otherwise it adds risk and cost without a matching benefit.

Why does the same carrier appear on both of our links?

Carriers often resell access over shared infrastructure, so two contracts can still depend on one duct, one exchange or one upstream network. Ask for the physical route of each circuit and for the carrier's own redundancy design, and record the answers in your documentation.

What is the most important thing to test?

A real failure with the application running. A laboratory test of link state does not show how a voice call, a cloud session or a payment terminal behaves during a switch. Schedule the test, inform the users, and record what each application did, including any session that had to be restarted.

About this content

General informational guide. Regulations and technical requirements change over time: always check the official sources and assess your specific case with a qualified technical or legal advisor.

GET IN TOUCH

Talk to our team.

We'll show how Altovar integrates into your infrastructure. No commitment, no sales pitch. Just a technical conversation.

BOOK A DEMO →