Your OLT Config Will Vanish: Backup, Change Windows and Rollback
The short version: most PON outages are not hardware failures — they are configuration losses. One deleted VLAN, one control board with no spare, one power blip that wiped an unsaved change, and every ONU in the building goes dark. A faster OLT will not save you. A backup list, a change window, and a three-tier rollback rule will.
1. Three Kinds of "Lost Config", Three Different Responses
When the site calls and says the OLT is dead, you are usually looking at three different problems wearing the same costume. Get the category wrong and your first move is wrong too — someone reboots, someone swaps a board, and the config that could still have been saved gets overwritten.
| Type | Typical trigger | Blast radius | First move |
|---|---|---|---|
| Human error | VLAN or service flow deleted, rate template pushed to the wrong profile, ONU authorisation cleared | One port or the whole chassis | Stop touching it; decide whether the action is reversible |
| Hardware failure | Control board, PSU, fan tray or flash storage dies | Whole chassis or frame | Find out whether an importable copy exists |
| Power loss / unsaved | Changes made but never written to flash; hard power cycle | Rolls back to the last save | Compare change time against last-save time |
| Version rollback | Firmware downgraded, config keys no longer recognised | Some service parameters lost | Confirm version-to-config compatibility first |
Only the second row actually needs spare parts. The other three are process failures — and process failures are more expensive, because the same team will repeat them.
2. What Actually Has to Be Backed Up on an OLT
Plenty of projects export one config file and call it done. Then the restore runs, and not a single ONT comes back — because ONU authorisation was never inside that file. Five asset classes need to be covered; skip one and you buy yourself two extra hours on site.
- Device configuration: the running config itself — uplinks, VLANs, service flows, alarm thresholds.
- Service templates: line and service profiles, DBA profiles, rate-limit templates, ONU capability profiles. Lose these and the config imports cleanly while nothing provisions.
- ONU authorisation and serial ledger: which serial numbers may register, which room each is bound to, which package tier it belongs to. Most overlooked, most expensive to rebuild.
- Uplink and addressing plan: VLAN-to-service mapping, LAG or link aggregation mode, gateway and route addresses, upstream handoff parameters.
- Version and integration details: firmware release, licence, NMS and northbound settings, SNMP community and read-only accounts.
There is a plain test for whether a backup is good enough: hand it to someone who has never worked on this site. If they cannot bring every service back within an hour, the package is incomplete.
3. Three Backup Methods and Where Each One Fails
All three methods in common use are useful, and none replaces another. Treat them as interchangeable and you have built yourself a very comfortable false sense of safety.
| Method | What it restores | Where it breaks |
|---|---|---|
| Manual local export | Full device config, complete rebuild | Needs a person on site; may not import after a version change |
| Scheduled NMS backup | Config for every site, high coverage | Depends on NMS and the link; a backup never tested is not a backup |
| Ledgers and template docs | Service relationships, room numbers, tiers | Cannot restore device parameters — rebuild only |
The combination that holds up in practice: NMS runs a nightly backup for coverage, engineers export manually at every milestone (new build, expansion, cutover, board swap) for usability, and the ledger keeps everything readable. Three sources that reconcile with each other — that is what "we have backups" should mean.
4. Change Windows: Four Rules That Are Not Up for Debate
Most incidents are not caused by a dangerous change. They are caused by the right change made at the wrong hour, by the wrong number of people, with no way back.
- Two-person rule. One operates, one reads back. This matters most on remote edits to a core port — reading the command out loud costs a minute; explaining it afterwards costs the night.
- Window selection. Hotels: 02:00–04:00. Residential: 03:00–05:00 — clear of checkout, settlement and the evening viewing peak. Campuses: weekends or after hours.
- Three things before you touch anything: export the config, screenshot the key pages, write the change ticket (what, why, rollback trigger).
- Thirty minutes after. Spot-check ONU online counts across three buildings, the uplink traffic curve, and one room each for voice and TV.
5. A Standard Change in Nine Steps
Fix the sequence and the human becomes a constant instead of a variable. Nine steps look fussy; each one is buying back time later.
- Write the change ticket: objective, scope of impact, rollback trigger, owner.
- Export the current config. Keep two copies — local and off-site — with date and version in the filename.
- Rehearse on an idle PON port or a lab unit. Confirm the commands behave as expected.
- Notify the site: front desk, engineering, on-call. State the expected interruption length.
- Execute one class of object at a time. Confirm it took effect before moving on.
- Save immediately — write the running config to flash.
- Watch for 15 to 30 minutes: online rate, alarms, optical power, uplink utilisation.
- Test real services: data, voice, TV, room control — one point each. Do not trust the dashboard alone.
- Close the ticket with actual duration and deviations. If a rollback trigger fired, roll back.
6. Rollback Matters More Than Repair: Three Tiers and Their Triggers
The classic failure mode on site is "let's just try one more thing". Three edits made, two services down, and everyone is still guessing which edit did it. Rollback is not surrender — it returns the system to a known state so you can close in on the problem instead of thrashing around it.
| Tier | Applies to | Cost | Decision deadline |
|---|---|---|---|
| Single service | One VLAN or service flow misconfigured | One service, seconds | Locate within 10 minutes |
| Single PON port | Batch ONU drop-off under one port | One port in one building | Not located within 10 minutes |
| Whole chassis | Impact beyond one building, or uplink involved | Site-wide interruption, announce it | Immediately, no waiting |
Two triggers, that is all: if you cannot localise it in ten minutes, step up a tier; if more than one building is affected, go straight to a full rollback. Put both lines in the on-call handbook — it is worth more than any post-mortem.
7. Board Swaps and Chassis Replacements: Moving the Config
A hardware replacement looks like unplug-and-plug. The real work is the config migration. A new board ships in default state and does not carry your service templates with it; a new chassis on a different model or release may not accept the import at all.
- Flash spare boards to the same firmware release as the live network, so imported parameters are actually recognised.
- Check whether slot and port mapping changed. If port IDs move, every service flow has to be re-mapped one by one.
- ONUs will re-register. This is not a power cycle — expect 30 to 90 seconds of interruption, and tell the residents in advance.
- After the swap: verify uplink and routing first, restore service templates second, confirm ONU authorisation in bulk last.
8. Three Ledgers That Must Exist
Configuration is written for the equipment. Ledgers are written for people. Without them, one staff departure turns the whole network into a black box.
- PON port map: port ID, splitter number and ratio, buildings and floors covered, uplink home.
- VLAN service map: VLAN ID, service carried, ports involved, upstream device and gateway.
- ONU ledger: serial number, room, package tier, install date, last replacement record.
The ledger does not need to be pretty, it needs to be inheritable. Reconcile it against the live config every quarter and fix the gaps on the spot — a ledger that disagrees with reality is the setup for the next outage.
9. Run a Fake Failure Every Quarter
An untested backup is emotional support, not disaster recovery. The drill needs no downtime — a spare chassis or a lab unit is enough. The point is to walk the process and measure the clock.
- Import the latest backup onto a unit of the same model.
- Confirm it auto-authorises ONUs by serial, that service templates are complete, and that uplink VLANs are reachable.
- Time the whole restore. Target: under 30 minutes.
- Feed every gap you find back into the checklist, the backup set and the ledgers.
10. Five Management Mistakes We See Most
- Backup stored next to the equipment. The rack burns, the backup burns with it. An off-site copy is not a formality.
- Config backed up, authorisation forgotten. The restore runs perfectly and not one ONT comes online.
- NMS backups never verified. Corrupt files, mismatched versions — you find out on the worst possible night.
- Changes not saved. Everyone assumes it took effect; the next power blip takes you back to yesterday.
- No change record. Three months later nobody can say who added that VLAN, or why.
Conclusion
PON reliability is half equipment and half human procedure. Hardware sets the ceiling; process sets the floor. Nail down backup, change window and rollback, and an ordinary network on ordinary hardware will feel remarkably solid. Skip them, and the best equipment on the market goes down to one careless edit. Before the next change, ask two questions: is the config exported, and where is the rollback point?
Contact Us
- Email: zhenghui.xu@sczhgx.com
- WhatsApp: +86 17612849559
- Online store: shop.sczhgx.com