Adding redundancy to a voice platform makes it more reliable, but it also makes the system harder to understand. Every backup path you add is another path that can fail without anyone noticing, and the backup that hasn’t been tested is often the one that lets you down during a real outage.
Geo-redundant SIP trunking means connecting your voice platform to your carrier from more than one geographic location, so that calls keep working if you lose a data center, a carrier point of presence or a network connection. A well-designed setup protects calls in both directions: outbound calls find another route, and inbound calls to your numbers reach a location that is still working. A poorly designed setup looks redundant on paper but fails as a single unit in practice, because its parts depend on something they share.
This article is written for UCaaS providers, contact center platforms and enterprises that run their own voice infrastructure. It explains how to design redundancy that holds up during a failure and how to test it regularly.
Start with what can fail
The best starting point for redundancy is a list of what can fail, rather than a list of equipment you can buy two of.
A failure domain is anything whose loss would bring down everything that depends on it. In a SIP trunking setup, the main failure domains are:
- your site, such as a data center, cloud region or office
- your network connection, such as an internet link or private circuit
- your session border controller (SBC), the device that handles call signaling and audio at the edge of your network
- the carrier’s point of presence, the location where your trunk connects to the carrier
- the carrier itself
- DNS, the records that tell each side where to send traffic
- certificates and credentials, because two systems with the same expiry date can fail on the same day
For each item on the list, ask which calls would stop if it disappeared. If the answer is “all of them,” you have a single point of failure, regardless of how many devices sit in front of it.
Hidden dependencies that defeat redundancy
The most common redundancy mistake is having two components that depend on the same thing without anyone realizing it. When that shared dependency fails, both components fail together, and the network diagram gives no hint of why.
For example, two SBCs in the same rack share the same power supply and network switches. Two trunks connected to the same carrier point of presence depend on that single location. Two internet connections from different providers can still run through the same cable duct into your building, and two cloud servers can sit in the same region. Even two SBCs whose certificates were issued on the same day will stop working on the same day if nobody renews them.
To avoid these shared dependencies, trace your connections down to the physical routes wherever possible. Place SBCs at different sites, connect each site to a different carrier point of presence, and renew certificates on different dates. If you also need protection against the carrier failing, you’ll need a second carrier, although that adds its own complexity in routing, number management and billing.
Outbound failover
Outbound failover is the more straightforward of the two directions, because your own SBC decides where each call goes.
To set it up, give each SBC more than one carrier destination, listed in order of preference. Then use SIP OPTIONS messages, which are small test messages sent at regular intervals, so the SBC knows which destinations are reachable before it sends a call. This avoids discovering a failure only when a call fails. When a destination stops responding to these messages, the SBC removes it from use, and when it starts responding again, the SBC brings it back.
Many setups also use DNS SRV records, which list several destinations with a priority and a weight for each. Systems try the destination with the highest priority first and move to the next one if it doesn’t respond. SRV records work well, but only if every system that uses them follows them correctly and the records are kept accurate. An out-of-date record that still points to a retired address can cause an outage as soon as traffic starts using it.
Inbound failover
Inbound failover is harder, because the decision is made inside the carrier’s network rather than your own. When someone calls one of your numbers, the carrier needs to know where to send that call and what to do if the destination doesn’t answer.
For this reason, inbound failover has to be configured explicitly with your carrier. Most carriers let you set an ordered list of destinations for each number or group of numbers, such as a main SBC, a backup SBC at another site, and a final fallback. You also need to agree what triggers failover, for example no response within a set time, certain SIP error responses, or both.
It’s worth planning a final fallback even if you hope never to use it. During a complete outage, forwarding your numbers to a mobile phone or an external answering service keeps your business reachable. If you manage numbers for many customers, store each number’s failover settings with the rest of its routing information. Our guide to multi-tenant DID inventory explains how to organize that information.
How quickly failures are detected
From a caller’s point of view, failover that takes 30 seconds doesn’t help, because most callers will have hung up by then.
The default timers in SIP are generous. If a call request gets no response, SIP waits for a timer known as Timer B to run out before giving up, and by default that timer lasts 32 seconds. If your SBC or the carrier waits for the full default before trying the next destination, callers hear silence for half a minute. You can avoid this by setting shorter timeouts for destinations you already monitor, and by sending OPTIONS messages often enough to detect a failure before calls are affected.
At the same time, you need to avoid switching back and forth too often. If a destination keeps failing and recovering every few seconds because of brief network problems, traffic can bounce between destinations and make the situation worse. To prevent this, require several missed OPTIONS messages before marking a destination as down, and several successful ones before bringing it back.
Active-active or active-passive?
There are two main ways to run two sites. In an active-active setup, both sites carry calls all the time. In an active-passive setup, one site carries all the calls while the other waits in reserve.
The main advantage of active-active is that it proves both sites work every day, because both handle real calls. Its main risk is capacity. If each site normally carries half the load, each one still needs to be able to carry all of it when the other fails, so each side should be sized for the full peak. Our guide to sizing SIP trunk channels explains how to calculate that peak.
Active-passive is simpler to manage, but the reserve site isn’t tested by real traffic until the day it’s needed. If you choose this approach, either send a small share of real calls to the reserve site at all times or test it on a regular schedule.
Making sure audio fails over too
Failing over the call setup is only part of the job, because the audio also needs a working path.
If an SBC fails while calls are in progress, those calls usually drop, since reconnecting live calls across sites is difficult and rarely worth the effort. What matters most is that new calls work straight away on the remaining path. To make sure they do, check that the audio paths, firewall rules and network address translation settings at the backup site are correct. A failover that connects calls without any audio is still an outage, and our guide to fixing one-way audio and NAT problems explains the usual causes.
Testing failover regularly
If a backup path hasn’t been tested recently, it’s safest to assume it doesn’t work. Configurations change over time, firewall rules are updated, and certificates expire on backup devices that nobody logs into.
To keep failover reliable, run scheduled tests that follow these steps:
- Announce a test window to the people who need to know.
- Deliberately take the main site or SBC out of service.
- Make inbound and outbound test calls, and record how long it takes until they succeed.
- On the backup path, check that audio works in both directions, that caller ID is correct and that transfers work.
- Bring the main site back and confirm that calls return to it without switching back and forth.
- Write down anything that failed and fix it before the next test.
For platforms that carry customer calls, testing once a month is a good starting point. Measure the time to failover on every test, and treat any slowdown as a problem to fix.
Writing a runbook before an outage happens
During an outage, nobody has time to read long documents carefully. It’s much better to write a short runbook while things are calm, and to store it somewhere that doesn’t depend on the systems that might fail.
The runbook should help your team answer a few questions quickly. How do we confirm which part has failed? Has traffic already moved automatically, or do we need to act? Who can change inbound routing with the carrier, and how, if we can’t reach our usual portal? What do we tell customers, and when? How do we move traffic back once the main site has recovered?
Include the carrier’s support contacts and escalation path, the steps for rerouting calls manually, and the list of test calls you use to confirm recovery. Review the runbook after every failover test and every real incident, because the design it describes will change over time.
Measuring availability as callers experience it
Knowing that your devices are switched on doesn’t tell you whether calls are working. A more useful measure is availability from the caller’s point of view: the share of inbound and outbound test calls that succeed with working audio, for each site and each path, over time. Automated test calls placed every few minutes give you this measure continuously, and they also catch problems on backup paths that normal traffic never uses.
Security as part of resilience
Attacks can cause outages in the same way that equipment failures do. For example, a flood of malicious traffic aimed at one SBC is easier to survive if calls can move to another one, and toll fraud through a compromised trunk may force you to take that trunk offline. Our SIP trunking security guide describes these threats and how to protect against them.
What to ask your carrier
Redundancy on your side only works if the carrier’s side supports it too. Ask your carrier how many points of presence you can connect to, whether inbound numbers can fail over between destinations and what triggers that failover, how quickly routing changes take effect, and how planned maintenance is announced.
IDT Express provides SIP trunking from 40+ locations, which allows platforms to connect each of their sites to a different point of presence so that both sites don’t depend on the same carrier location. If you use Microsoft Teams, our guide to what breaks in Teams Direct Routing covers SBC health checks on that side, and our SIP trunking page describes the service.
Frequently asked questions
What is geo-redundant SIP trunking?
Geo-redundant SIP trunking connects your voice platform to the carrier from more than one geographic location, so the loss of a site, network connection or carrier point of presence doesn’t stop calls. It should protect both outbound and inbound calls.
How does inbound SIP trunk failover work?
The carrier routes calls to your numbers using an ordered list of destinations. If the main destination doesn’t respond within a set time or returns certain errors, the carrier sends the call to the next destination on the list.
Why does SIP failover take so long?
By default, SIP waits 32 seconds for a response before giving up on a call request, so a system that relies on the default leaves callers in silence. Shorter timeouts and regular SIP OPTIONS checks detect failures much faster.
Is active-active better than active-passive for SIP trunks?
Active-active proves both sites work every day, but each site must be sized to carry the full load on its own. Active-passive is simpler, but the reserve site needs regular testing because it doesn’t carry real traffic.
How often should SIP failover be tested?
Once a month is a sensible starting point for platforms that carry customer calls. Each test should measure how long failover takes and check inbound calls, outbound calls, audio and caller ID on the backup path.
Talk to our SIP team
If you’re designing your voice platform to survive the loss of a site, talk to the IDT Express SIP team about connecting your sites to separate points of presence with inbound failover.


