The full power of TDengine, now free forever for up to 5,000 tags.

Explore More

Integrating Historians Across Multiple Sites: Architecture Decisions and Tradeoffs

Juno Qiu

November 3, 2026 /

Two plants can produce the same product and still have very different histories. One may use a long-established historian with years of process data; another may use a different platform, naming convention and network design. A central reliability team may want to compare equipment across both sites, while each plant must continue to support local operations.

Multi-site historian integration makes selected data available across those boundaries. The architecture must answer more than where data will be stored. It must establish who owns each measurement, which queries or copies are permitted, what a shared result means when a site is unavailable and who is responsible for recovery.

An integration project does not automatically require replacing the site historians. Begin with the cross-site decision that needs data, then select the least disruptive access and storage arrangements that satisfy it.

Start with one cross-site decision

Suppose a reliability team wants to compare weekly pump operating conditions at two plants. Site A holds pressure history in its existing historian. Site B has another historian and stores a similar measurement under a different tag and unit. Each site uses its own history for local investigations.

The team needs a defined set of measurements, the time range and resolution needed for the comparison, acceptable delay, and a rule for identifying comparable pumps. The sites must approve interfaces and source load. They also need to agree whether the central result can be shown when one site is delayed or missing.

This is an illustrative example, not a prescribed deployment. Actual paths may include site collectors, access services, intermediaries or other approved components. Neither the central application nor a shared store needs unrestricted access to plant controllers.

Create a site inventory before choosing a platform:

QuestionDecision to record for each site
What is the source?Historian and owner, available interface, measurements and historical range
What is shared?Approved tags or derived values, units, quality, resolution, retention and intended use
How is it reached?Authorized network path, query or collection limits and credentials
Who operates it?Site, integration and destination owners, support hours and escalation path
What happens on failure?Detectable delay, available recovery, acceptable data age and handling of missing results

Do not assume that the same connector, permissions or history are available at both sites. A shared architecture may require site-specific implementation and acceptance criteria.

Decide where cross-site queries are served

The basic choices are to query approved site sources when needed, copy a selected dataset to a shared destination, or combine the two. These are approaches to access and storage, not mutually exclusive products or an exhaustive taxonomy.

Cross-site decisionWhat the architect needs to agree
Site independenceIdentify which local historian services and investigations must continue if the central service is unavailable. Each site approves the interface, workload and maintenance window it can support.
Partial or missing dataDefine the maximum acceptable data age for each site and whether a comparison is suppressed or shown with the missing or delayed site clearly identified. A previously copied record may remain available without being current.
Failure isolationDecide whether a site or network failure should affect only that site’s contribution or the whole cross-site query. Test the actual dependencies: a shared destination or common integration service can itself be a single point of failure.
Responsibility boundaryAssign site owners for source access and load, an integration owner for transfer and recovery, and a destination owner for shared availability. Agree who alerts consumers and resolves discrepancies across those boundaries.
Changes and recoveryDefine how each site exposes corrections, deletions and missing history, and whether the chosen path can carry them. Assign an owner to verify recovery and decide when affected comparisons must be recalculated.

For the pump comparison, a shared copy of a limited set of measurements may be reasonable if the central team runs the same reports frequently and each site accepts the collection load. Detailed troubleshooting history could remain at each site for authorized, occasional access. This choice is conditional: validate the real interfaces, network design, site constraints and support capacity before implementing it.

An on-demand design may fit when cross-site requests are infrequent and source interfaces can safely handle them. It reduces the data copied but makes each request dependent on the source and network at that time. A shared copy can isolate repeated central queries from the source, but creates a separate dataset that needs freshness, quality and recovery controls. A hybrid design inherits responsibilities from both.

Give every shared measurement an owner and a meaning

A central name such as pump_pressure is not enough to make Site A and Site B comparable. Confirm measurement location, unit, operating range, sampling or compression behavior, time convention and quality representation. A pressure recorded upstream of a pump is not interchangeable with one recorded downstream simply because both have similar names.

Keep the original site and source identifier alongside the common asset identity. Define who approves a mapping, who changes it after equipment replacement and how the effective date is recorded. If a unit is converted, keep the conversion rule and precision. Where the source exposes quality or other context, decide what the destination must preserve; verify actual connector behavior instead of assuming all metadata travels with the value.

The site historian can remain the source of record even when selected data is copied elsewhere. Write down which system users should consult for a local incident, which dataset supports the cross-site comparison and how a discrepancy is escalated. Industrial data contextualization provides additional background on connecting measurements with equipment context.

Design for a missing or delayed site

Assume Site B’s connection to the shared destination fails for four hours while Site A continues sending data. A dashboard that quietly compares the latest data from A with stale data from B can produce a plausible but misleading result.

Define the response before an outage:

  • Detection: Which team receives an alert, and how is freshness measured for each site and dataset?
  • Presentation: Does the consumer suppress the comparison, show a partial result with the missing site identified, or display the age and completeness of each contribution?
  • Recovery: Which component, if any, buffers data? What are its capacity and retention limits? Can the source provide missing history later, and who initiates and verifies recovery?
  • Recalculation: If late data changes a weekly total or comparison, who decides whether and when to recompute it and notify users?

A buffer protects only data that reaches it, and its existence does not prove that every missing source record can be recovered. Test outage duration, storage capacity, replay, duplicates and reconciliation against the actual components. If a gap cannot be reconstructed, retain it as a known limitation rather than filling it with an assumed value.

Assign responsibilities across the boundary

The site owner approves the source interface, local load and maintenance windows; the integration owner monitors collection and recovery; the destination owner manages the shared service and data access; and the consuming team defines how stale or missing data affects its decision. One organization may hold several roles, but the handoffs should be explicit.

Review network zones, access rights, credentials and any intermediary with OT and security owners at each site. Do not assume a single enterprise design can cross every site’s boundary unchanged. NIST’s OT security guide provides a risk-based starting point for balancing security with operational reliability, performance and safety.

Document support windows and escalation as well as technical ownership. If a connection fails on a weekend, the agreed service level should say who responds, whether collection can wait and how consumers learn that the shared dataset is incomplete.

Pilot two sites before standardizing

Use one bounded comparison and a small number of approved measurements at each site. Include at least one difference in units, naming or source behavior so the pilot tests real integration work rather than only connectivity.

  1. Confirm the site owners, source interfaces, permitted load and data use.
  2. Map selected measurements and verify their meaning, units, timestamps and quality with each site.
  3. Test the required history and current updates separately; compare selected source records with what the consumer receives.
  4. With site approval, exercise a defined site or network interruption. Verify alerting, partial-result behavior, recovery and any recalculation.
  5. Review measured freshness, source impact, exceptions and support ownership against agreed acceptance criteria.

Scale only after the teams can explain what the shared result includes, which site data is missing or delayed, and who owns each unresolved exception. The number of connected historians is less useful than a cross-site result that engineers can interpret and operate with confidence.

Related Articles