Migration
PI stays up. We rebuild the model.
There is no weekend cutover. LCT keeps PI live, streams and backfills in parallel, and rebuilds the asset model and knowledge on TDengine. Applications move when parity is proven — and PI is still there if you need to roll back.
We migrate the plant, not just the tags
Copying points is not the job. LCT stands TDengine up beside PI, rebuilds the asset model your AF tree only implied, and attaches the manuals and failure history the new historian will actually need. PI keeps serving the plant the whole time. Rollback is the default posture: both historians run until you choose to retire the old one.
Architecture notes that matter
- taosX reads PI so the old historian never has to go down first. Live tasks stream new values; backfill tasks pull history by time range.
- The connector uses the PI AF SDK on Windows: run taosX next to PI, or keep TDengine elsewhere and place a Windows taosX-Agent on the PI network.
- PI Data Archive–only sources typically map to one tag per TDengine table — the single-column model.
- PI Data Archive + AF Server paths can also map an AF element and its attributes into one table per element — the multi-column model — via the PI AF SDK.
- TDengine documents two migration orders. Backfill first, then real-time, is their usual recommendation. Real-time first, then backfill, is valid when live data cannot wait — overlapping timestamps overwrite rather than duplicate.
- Backfill increases load on PI — split by template/point groups and throttle batches.
Migration steps
1. Assess the PI system
Inventory PI Data Archive points, AF structure, users, interfaces, and dependent applications. Identify what must migrate first versus what can wait.
2. Deploy TDengine TSDB and IDMP
Stand up TDengine Historian components in the target environment (on-prem or approved cloud) and confirm networking, auth, and backup baselines.
3. Configure the PI connector path
Use TDengine taosX with the PI connector (PI AF SDK on Windows). If taosX is not on a Windows host that can reach PI, deploy a Windows taosX-Agent. Data Archive-only sources map to single-column models; Data Archive plus AF Server can use single- or multi-column mapping.
4. Start real-time streaming
Create a PI real-time task so new values flow into TDengine while the existing PI System remains operational.
5. Backfill history in controlled batches
Run PI backfill tasks by time range and point groups, monitoring PI load. Validate counts and spot-check values before expanding scope.
6. Rebuild context, dashboards and cut over
Model assets in IDMP, recreate critical panels and notifications, then phase applications off PI once parity is proven.
Validation checklist
- ✓Compare point/tag counts for the same time range
- ✓Spot-check values and timestamps (watch time zones)
- ✓Confirm critical dashboards and notifications in IDMP
- ✓Only then retire dependent PI client workflows
Migration questions
How long does a PI to TDengine migration take?
It depends on how many tags, how much history, and how many applications read from PI. You do not wait until the end to see value: real-time streaming into TDengine can be running early, so operations can already query and chart the new historian while backfill and cutover continue in batches.
Does PI stay running during the migration?
Yes. PI stays live and remains the system of record until you choose otherwise. TDengine fills beside it. Applications move only after we have matched counts and spot-checked values, so there is always a way back.
What happens to PI Asset Framework models?
We inventory the AF structure in assessment and rebuild it in TDengine IDMP — templates, attributes and the relationships your team actually uses. The taosX PI connector (PI AF SDK on Windows) maps PI Points or AF elements into TDengine tables; that is ingest into TSDB, not a one-click copy of AF into IDMP. Real-time and backfill tasks run while PI stays live, so you do not need a weekend cutover for the data path.
Get a migration sketch for your PI system
Tell us the tag count, how much history has to come across, and which applications read from PI. You get a phased sketch: what streams first, what backfills, what moves last.