Why Configuration Backups Belong Next to Monitoring
Monitoring tells you that a switch is down. It does not tell you what was on it. The moment the hardware fails for good, the question changes from "is it up?" to "what exactly did it do?" That means which VLANs, which trunks, which port carried the phone system. If the only copy of that answer lived in the dead device's flash memory, you are now rebuilding a switch from memory. You are relying on a wiring diagram last accurate before you were hired, and whatever a colleague can recall about port 17.
A configuration backup is a text file. It contains the settings that make a device your device rather than a factory-fresh one. Those settings are copied off the device and stored somewhere safe. Yet that one small file does three different jobs. It is your disaster recovery plan. It is the most honest documentation you will ever own. And, compared with yesterday's copy, it is the cheapest change detector you will ever run.
This guide is the "why" of the topic. By the end you will know what a configuration is and what the three payoffs really buy you. You will know which devices need backing up and roughly how, how often, and where to keep the copies. You will also know why the backup store is itself sensitive data. And you will know why a backup you have never restored is closer to a hope than a plan. The sibling guides cover the automation, the diffing, and the workflow in depth. This one gives you the reasons to bother.
In this guide
- What a configuration actually is
- Payoff one: disaster recovery
- Payoff two: documentation that cannot lie
- Payoff three: the cheapest change detector
- What to back up
- How often
- Where to keep it
- Secrets inside configs
- Restore testing
- The inherited network
- Where to go next
What a configuration actually is
Every managed network device ships with a default personality and then gets customized. The customization is the configuration. It includes hostname, management address, VLANs, what each port does, routes, and access lists. It also includes where it sends logs, which time servers it trusts, which users may log in, and its SNMP settings. On most devices you can print the whole thing to the screen with one command from the show running-config family. What comes back is plain text, line after line, readable by a human:
hostname sw-floor2-01 ! vlan 30 name guest-wifi ! interface GigabitEthernet1/0/17 description Phone system gateway switchport access vlan 50 ! logging host 10.0.5.20 ntp server 10.0.5.21 snmp-server community R3adOnly-Lab RO
Two copies of that text usually exist on the device. The running configuration is what is in effect right now. The startup configuration is what the device will load the next time it boots. They can disagree: someone makes a change at the CLI, tests it, walks away, and never saves. The device works fine for months, then a power blip reboots it and the unsaved change evaporates. A backup should capture the running config, because that is the truth of what the network is doing. A good backup job also notices when running and startup differ. That gap is a reboot-shaped outage waiting to happen.
Not every device speaks in plain text. Some firewalls, wireless controllers, and load balancers export their configuration as a structured file or an opaque archive. The principle is identical: a file that, loaded onto a replacement unit, reproduces the device. Whether you can read it in a text editor matters for the diffing payoff, less so for the other two.
Payoff one: disaster recovery
Picture the worst Tuesday. The distribution switch for the second floor dies — not a reboot, a genuine hardware death. You have a spare of the same model in the storeroom. Now the whole question is whether you have its configuration.
Without a backup, you are reconstructing. Which VLANs were trunked upstairs? What were the uplink settings? Which ports fed the access points, and were they on a different VLAN from the printers? Every gap you fill from memory is a place where the rebuilt switch will behave slightly differently from the old one. And each of those places is a ticket you will receive over the following week.
With a backup, the sequence is almost boring. Unbox the spare. Connect a console cable. Load the file — often simply a paste at the CLI. Check that the interface names match the new hardware. Save. Cable it in. A dead switch is a spare plus a config file plus perhaps an hour of your time. That is the entire disaster recovery plan for that device, and it costs a nightly script to maintain.
The same file also rescues you from the disaster you cause yourself. A change goes wrong, the device is alive but doing the wrong thing, and the fastest way back is to load last night's configuration. Rollback and restore are the same operation with different emotions attached. One caveat: a configuration saved from one firmware release may not load cleanly onto a device running a very different one. The guide on safe firmware upgrades covers keeping the two in step.
Payoff two: documentation that cannot lie
Your network diagram says VLAN 30 is the guest wireless network. The configuration on the core switch says what VLAN 30 actually is, right now, with no possibility of being out of date. Documentation written by people drifts from reality the moment it is finished; the configuration is reality. A folder of last night's backups is a complete and current description of the network that nobody had to remember to update.
This turns a category of slow questions into fast ones. "Which devices still reference the old syslog server?" becomes a search across the backup folder rather than a login to forty devices. "Does anything still have the default SNMP community?" — search, and then a slightly uncomfortable afternoon. "What was on port 12 before it was repurposed?" — open the older copy. None of this touches a live device, so it is safe to do at two in the afternoon on a weekday.
What the configuration cannot tell you is why. It says port 12 is in VLAN 40. It does not say that VLAN 40 exists because of a compliance requirement rather than a technical one. The reasons live in the documentation around the config. And the guide on documentation that survives is the place to learn how to keep that layer honest.
Payoff three: the cheapest change detector
A diff is a comparison of two text files that reports only the lines that differ. It is a standard tool on every operating system. Applied to two backups of the same device, it answers one question with perfect reliability: did anything change? If today's file matches yesterday's, nothing changed. If it does not, something did, and the diff shows you exactly which lines.
Think about what that gives you. Every configuration change on the network — planned, emergency, or made by a contractor who never mentioned it — produces a difference between two backups. You do not need anyone to file a form, and you do not need to trust anybody's memory. You need two files and a comparison. When the comparison finds a change you can match to a change record, that is fine. When it finds one nobody announced, that is drift. Drift is where a surprising share of "the network is being weird" tickets come from.
This payoff is why configuration backups belong next to monitoring rather than in a separate mental drawer. Monitoring watches metrics: is the port up, how busy is the link, how hot is the chassis. Backups plus diffs watch configuration: did someone change what the port is supposed to do. An alert that says "interface down" and a diff that says "someone shut that interface at 09:14" are two halves of the same story. The mechanics are in the sibling guide on diffing configs and alerting on changes nobody announced.
What to back up
The honest answer is "anything with a configuration you would hate to retype." Start with the devices whose failure would take other things down with them, then widen the circle.
- Switches — core, distribution, and access. Access switches feel unimportant individually. There are also the most of them, and each carries a unique set of port assignments written down nowhere else.
- Routers and anything else doing routing, including layer-3 switches and site-to-site VPN endpoints.
- Firewalls — rules, objects, NAT, VPN definitions. Often the most complex configuration on the network and the costliest to get wrong.
- Wireless controllers, and standalone access points if they hold their own configuration.
- Load balancers and reverse proxies, whose configuration is often the only record of how an application is actually wired.
- Out-of-band gear — console servers and management switches, the things you use to fix everything else.
- The monitoring system's own configuration — device lists, thresholds, alert rules, dashboards. Losing it means losing your eyes at exactly the moment you need them. The guide on backing up the monitoring system itself covers that case.
A useful way to organize the job is by how each device type hands over its configuration. The table below is a starting point, not a rule; adjust it to what your devices actually support.
| Device type | Typical backup method | Suggested frequency | Notes |
|---|---|---|---|
| Access switch | SSH login, show running-config, save output | Daily, plus after any change | Many devices; port descriptions change often |
| Core / distribution switch | SSH pull, or device-initiated copy to a file server | Daily, plus after any change | Highest blast radius; verify running matches startup |
| Router / VPN endpoint | SSH pull | Daily, plus after any change | Contains pre-shared keys |
| Firewall | Management API or export feature; sometimes SSH | Daily, plus after any change | Export may be structured, not plain text |
| Wireless controller | API or export feature | Daily | Contains wireless pre-shared keys |
| Load balancer / proxy | API or export; sometimes a file copy | Daily | May include certificate private keys |
| Console / OOB devices | SSH pull or export | Weekly is often enough | Rarely changes; painful to rebuild blind |
| Monitoring system | Its own export, plus database backup | Daily | See the monitoring-backups guide |
How often
Daily is a fine default. A nightly pull is cheap for the devices (one login, one command), and cheap for you (one scheduled job). It gives you a clean rhythm: today's file versus yesterday's, one diff per device per day. If the network changes on Wednesday, Thursday morning's diff shows it. That cadence is enough for most small and mid-sized networks and it is where nearly everyone starts.
After every change is better. A daily backup captures one moment; if two changes happen between pulls, the diff shows both blended together. A backup immediately after each planned change gives you a before-and-after pair per change. That is exactly what you want when verifying that the change did what it was supposed to and nothing else. The practical arrangement is both. A nightly job acts as the safety net. A hook in the change process triggers an extra pull right after a change. The sibling guide on a change workflow that keeps backups, diffs, and alerts in sync shows where the post-change pull fits.
Could you go hourly? On a large network with many hands making changes it can be worth it; for most teams it mostly produces identical files. Increase the frequency only if you find yourself wishing you had a copy from three hours ago.
Where to keep it
The backups need a home that does not share fate with the devices, is access-controlled, and keeps history rather than overwriting the previous copy.
Not the same box. Some devices offer an archive feature that keeps the copy on local flash. A backup stored only on the device itself is not a backup; it dies with the device. A backup stored only on a virtual machine behind the core switch is better. But if the core switch is the thing that died, the backups are on the far side of the outage. Keep the primary store somewhere you can reach with the network partly broken. And keep a second copy somewhere else entirely — another site, or cloud storage that only the backup account can write to.
A version-controlled repository. Version control keeps every historical version of a set of files along with who changed each one and when. The tool most people reach for is git, built for source code but equally happy with configuration text. Storing backups in a repository gives you the diff payoff for free, an unlimited history, and a log of exactly when each device's configuration changed. A repository of a few hundred device configs with a long history is tiny; there is no storage argument against keeping everything.
Access-controlled. Read on, because this one deserves its own section.
Secrets inside configs
Open any device configuration and count the things you would not want a stranger to read. Local user passwords, usually hashed, but on some platforms with hashes weak enough to be reversed in seconds. SNMP community strings. Pre-shared keys for site-to-site VPNs. The shared secret between the device and your authentication server. Wireless pre-shared keys. API tokens. On some appliances, the private keys of TLS certificates. A configuration file is, among other things, a neat list of ways to get into your network.
That makes the backup repository sensitive data — at least as sensitive as the devices themselves. It is arguably more sensitive, because it is all in one place. Treat it that way:
- Restrict who can read it to the people who could log into the devices anyway. It should not be world-readable on a file share, and never in a public or organization-wide repository on a hosting service.
- Encrypt it at rest, especially the off-site copy.
- Do not paste configs into tickets, chat, or email without stripping the secret lines first. A diff alert should include the changed lines, but the script should mask anything that looks like a key or password before sending.
- Keep the backup credentials out of the scripts. The account that pulls configs should be read-only where the platform allows it. And its password or key should come from a protected store at run time, never from a line in the script.
- Decide about redaction deliberately. Some teams strip known secret patterns on the way into the repository. That makes the store safer if it leaks, at the cost that a restore then needs the secrets from somewhere else. If you redact, keep an unredacted encrypted copy that only the restore process can reach.
Tip: consider the first time you search the backup folder for public. You find the default SNMP community on a device you had forgotten existed. You have learned two things: why backups are documentation, and why the folder needs a lock on it.
Restore testing
A backup you have never restored is a hope. It might be a well-founded hope — the file looks right and contains the hostname. But until a device has actually booted from it and behaved correctly, you do not know. The things that go wrong at restore time are rarely visible in the file.
The common surprises: the spare is a different hardware revision with differently named interfaces, so half the port configuration fails to apply. The configuration was saved from a newer firmware and uses commands the spare does not recognize. A password hash tied to the hardware does not transfer, and the restored device has no working admin login. A certificate referenced by the config is not in the config, and the management interface refuses to start. None of these are catastrophic on a lab bench. All of them are miserable at two in the morning with the second floor offline.
So test. Keep one spare of each critical model, or borrow a device during a quiet period. And a few times a year run the drill. Pick a recent backup and load it onto the spare. Confirm it boots, confirm the management address responds, and confirm a couple of representative settings are in effect. Write down the steps and the gotchas as you go. That document becomes the restore runbook, and it is worth more than the backup file itself. If the restore needs manual fixes, feed those back into the backup process. Perhaps the job should also capture the firmware version and hardware model alongside the config. That way, the person restoring knows what to match. A backup job without a restore test is only half a job.
The inherited network
Many administrators meet this topic by inheritance. They find a new job, a departed predecessor, a network with no backups, no documentation, and a drawer of unlabeled console cables. It feels like the worst possible starting point. It is actually the best argument for backups, because the first useful thing you can do with an unknown network is to capture it.
Log into each device you can find, pull its configuration, and save it. Do not try to understand everything yet. Just collect. Within a day or two you will have a folder that is, for the first time, a complete description of what the network is. That description is searchable, diffable, and safe from the next hardware failure. Then the reading begins. The guide on auditing inherited configurations walks through what to look for once the files are in hand. Collecting also surfaces devices you did not know about: a configuration references a neighbor by address. The neighbor is not on your list, and now it is. Cross-checking what the backup job found against what the monitoring system discovered makes both lists more complete.
Where to go next
A configuration backup is a text file that does three jobs. It rebuilds a dead device from a spare in an hour instead of a week. It documents the network with a truthfulness no hand-written page can match. And compared with the previous copy, it detects every change whether or not anyone announced it. To get those payoffs, back up every device with a configuration you would hate to retype. Do it daily at minimum and after every change ideally. Use an access-controlled, version-controlled repository that does not share fate with the devices. And restore from it at least once so that you know it works.
The rest of this topic turns those principles into practice. The guide on automating config backups covers pull methods, scheduling, and repository layout. The guide on diffing configs turns the daily comparison into an alert you can trust. And golden configs define what a correctly configured device of each role looks like. And the change workflow ties it to the change process so that every drift alert means something. All of them live under Config Backup and Drift Detection Alongside Monitoring.
Put this into practice with SysPeep Monitor
Configuration backup is a companion practice to monitoring rather than a monitoring feature. SysPeep Monitor does not claim to do it. What SysPeep does is watch the devices those configurations live on. It discovers them and collects health and performance metrics over SNMP, SSH, and similar collectors. It raises alerts on thresholds and outages. That way, when a restore is needed, you learn about the failure from an alert rather than a phone call. The free trial installs on a Windows machine in minutes.
Download Free TrialLearn About SysPeep