OSIsoft PI

What is OSIsoft PI?

OSIsoft PI, now the AVEVA PI System, is an industrial data platform and historian that collects process data from PLCs, DCSs and SCADA systems in real time, stores it efficiently as time series and makes it available in context for analysis and IT systems. The platform was developed by OSIsoft, founded in 1980 by J. Patrick Kennedy in San Leandro, California. AVEVA acquired OSIsoft for around 5 billion US dollars; the deal closed on 19 March 2021, and since then the PI System has been part of AVEVA, which has been wholly owned by Schneider Electric since January 2023. In the process, energy, water and pharmaceutical industries PI is the de facto standard for enterprise historians, and it often forms the central data layer in OT convergence.

🧠 What does the PI System do?

The PI System has four core tasks:

• Collect - reads measured values (tags, called PI Points in PI terms) from controllers, sensors and systems through interfaces, connectors and adapters
• Store - keeps years of process data as time series in the PI Data Archive, using filtering that discards noise but preserves the process behaviour
• Contextualise - links tags to assets such as pumps, reactors and production lines through the Asset Framework
• Deliver - serves data to operators, engineers, data analysts and IT systems such as MES, ERP and cloud platforms

A PI installation can range from a few thousand to millions of tags, spread across dozens of sites. PI is therefore not a replacement for SCADA or a DCS: it controls nothing, but it remembers and exposes what those systems do.

🔧 What are the components of the PI System?

Formally, the PI Server consists of the PI Data Archive and the PI AF Server; the other components build on top of it. Modern AVEVA adapters send data using the Open Message Format (OMF) to a PI Server, to AVEVA Data Hub or to Edge Data Store.

⚙️ How does PI store so much data efficiently?

PI filters data in two stages before it is stored permanently.

1. Exception reporting - the interface or adapter only forwards a new value if it differs from the previously sent value by more than the exception deviation (ExcDev), or if the ExcMax time has elapsed. This is a simple deadband that mainly saves network traffic.
2. Compression - the PI Data Archive compares each new snapshot with the last archived point using the swinging door algorithm. If intermediate values lie on a straight line within the compression deviation (CompDev), they are not stored; PI interpolates between stored points when data is retrieved.

AVEVA’s rule of thumb is to set CompDev to the precision of the instrument (or lower) and ExcDev to half of that. In a field study AVEVA presented in 2023, exception reporting alone cut network traffic by about 70%, and the combination of exception and compression cut stored data by about 94%. The downside: settings that are too wide flatten peaks, so averages and trends reflect the algorithm rather than the real process. For critical measurements, such as data under GxP or 21 CFR Part 11, administrators therefore deliberately choose tighter settings or no compression at all.

🏭 Where does PI sit in the OT architecture?

In the Purdue Model, PI does not live at level 1 or 2, although it collects its data there.

A common pattern is one PI Server per site at level 3 and a second PI Server in the DMZ or IDMZ that is fed by replication (for example PI-to-PI). Office users then read only the copy, so no IT connection ever enters the production network. For stricter separation, a data diode can be placed in that replication flow. The Asset Framework maps well onto the hierarchical equipment model of ISA-95.

📡 How does data get into PI?

PI offers hundreds of collectors. The main sources and protocols are:

• OPC UA, OPC DA and OPC HDA - the most common route to PLCs, DCSs and SCADA servers; IP devices are usually reached through an OPC server
• Modbus and Modbus TCP - for meters, pumps and simple controllers
• DNP3 and MQTT - through AVEVA adapters, for telemetry and IIoT devices among others
• Files and databases - structured files and relational sources, for example from laboratory systems

Every collector buffers locally. If the connection to the PI Server drops, the PI Buffer Subsystem holds the data and forwards it later. With n-way buffering a collector writes simultaneously to all members of a PI Data Archive collective, a group of redundant servers that thus all receive the same data and can take over from one another on failure.

📊 What are Asset Framework and Event Frames?

The Asset Framework (AF) turns raw tag names such as FIC101.PV into recognisable objects: “Pump P-101 > flow”. With templates you define once what a pump looks like (attributes, units, calculations) and apply it to hundreds of pumps. Dashboards and analyses then work on the template instead of individual tags.

Event Frames record time periods with a start, an end and context: a batch, a fault, a start-up or an alarm period. That makes comparisons such as “all batches of product X in 2025” possible, which fits batch standards such as ISA-88 and OEE analyses. The PI Analysis Service creates Event Frames automatically from a start and end condition, for example “flow above 0” for a running pump.

🔐 How do you secure a PI System?

A PI System holds the entire process history and has connections to both OT and IT. That makes it an attractive target and a potential stepping stone. CISA advisory ICSA-24-018-01 (January 2024), for instance, described two vulnerabilities in PI Server (CVE-2023-34348, CVSS 7.5, and CVE-2023-31274) that let an unauthenticated attacker crash the PI Message Subsystem or exhaust its memory. Later advisories in 2025 and 2026 covered PI Server, PI Data Archive and PI Web API.

Document these measures in a zones-and-conduits design following IEC 62443 and NIST SP 800-82, as part of Defense in Depth and a Zero Trust approach. For Dutch organisations covered by the Cybersecurity Act (Cyberbeveiligingswet), in force since 15 August 2026, the historian is among the systems to which the duty of care and incident reporting apply; ISO 27001 and the NIST CSF provide the management framework.

🧭 How do you approach a PI implementation or optimisation?

1. Inventory sources and tags - which systems, protocols and scan rates are really needed?
2. Set exception and compression per measurement type - record CompDev, ExcDev and scan rate in a master table per unit
3. Design the AF model first - templates per asset type, aligned with ISA-95 levels
4. Choose the architecture - site PI at level 3, replication to the IDMZ, buffering on every collector
5. Set up security and availability - collective, backups, AD groups and segmented ports
6. Plan the lifecycle - versions, interfaces migrating to adapters and certificates belong in lifecycle management

Upgrades in production environments require validation and a test environment, because reports, analyses and links to MES depend on the PI System.

🔄 How does PI compare with other historians?

PI stands out mainly for scalability, the Asset Framework and its broad ecosystem of integrations. Open alternatives are cheaper but require more of your own work on collection, context and administration. Increasingly, PI also publishes data to Apache Kafka, a Unified Namespace or a digital twin, and supplies the history for predictive maintenance and edge applications.

❓ Frequently asked questions

Is OSIsoft PI the same as the AVEVA PI System?

Yes, the AVEVA PI System is the same product as OSIsoft PI. AVEVA closed its acquisition of OSIsoft on 19 March 2021 and then introduced the AVEVA name. Many users and documents still refer to it as OSIsoft PI.

What is PI Vision?

PI Vision is the web-based visualisation tool of the PI System, letting operators and engineers view dashboards and trends in a browser. PI Vision works on Asset Framework templates, so one display can serve every similar asset. It is the successor to the older Windows application PI ProcessBook.

How does compression work in PI?

PI first applies exception reporting as a deadband at the collector and then the swinging door algorithm in the PI Data Archive. Values that lie on a straight line within the compression deviation are not stored and are interpolated on retrieval. AVEVA recommends setting CompDev to the instrument precision and ExcDev to half of that.

Is the PI System a SCADA system?

No, the PI System is a historian and data platform, not SCADA. SCADA and DCS systems control and supervise the process, while PI collects their data, stores it long term and exposes it in context. PI can write values back, but in practice it is rarely used for control.

Where should you place a PI Server in the network?

A site PI Server usually sits at level 3 of the Purdue Model, close to the process sources. Office users should preferably read a replicated PI Server or PI Web API in the IDMZ, so no direct connection from IT into the production network is needed. A data diode in the replication flow adds extra assurance.

What are the alternatives to PI?

Alternatives to the PI System include AVEVA Historian, historians from DCS vendors and time-series databases such as InfluxDB or TimescaleDB. These are sometimes cheaper or simpler, but often lack the Asset Framework and the broad range of OT collectors. The right choice depends on scale, number of sites and the context you need.

📌 In summary

OSIsoft PI, now the AVEVA PI System, is the enterprise historian that collects process data, compresses it efficiently and makes it available in context for OT and IT through the Asset Framework. Its value lies in scale, reliable buffering and asset context; its risk lies in its central position between the production network and the office. Sound compression settings, an IDMZ architecture and security following IEC 62443 make PI a solid foundation for IT/OT convergence.