Data connectivity

CONNECT supports scalable high-throughput data ingestion and egress across varied deployment environments, from edge devices to enterprise-grade servers.

This page provides an overview of the performance benchmarks and deployment considerations relevant to key connectivity components.

Adapters

Adapters are installed on a variety of Linux and Windows devices and provide flexible data ingestion options for both CONNECT and PI Server environments. Supported hardware for adapters ranges from Raspberry PI to large Windows servers.

Test bed description

• Tested on a Windows desktop machine (4-core CPU, 4GB RAM).

Performance metrics applied to all Adapters

Additional notes:

• The actual performance of each Adapter will vary based on the data shape, the performance of the data source, the resources available on the host device, and/or the network bandwidth.

• In the events/second test metric, an event is defined as a timestamp, a data value and an optional quality value. A quality value is sent with the timestamp and data value if provided by the data source.

• The 50,000 stream count and 50,000 events/sec figures are test targets and not a failure threshold.

• Performance metrics apply to CONNECT and PI Servers as the data destination.

Edge Data Store

Edge data store (EDS) is optimized for edge deployments with real-time data ingestion and periodic egress to downstream systems. Edge Data Store is installed on a variety of Linux and Windows devices from Raspberry PI to large Windows servers.

Test bed description

• Tested on a Windows desktop machine (4-core CPU, 4GB RAM).

• Tested to the current commercial entitlement limits.

• Tests were performed to equalize inbound and outbound data rate at 1 event per second, per stream.

Performance metrics

Additional notes:

• The test cases involved outbound data sent from Edge Data Store as it arrived, at 1 event/sec/stream. However, in actual use it is more common that inbound data resides in the data store for a period of time and then the data collected in the data store is sent on a schedule, for example once an hour, at the end of a shift, or once a day.

• In the events/second test metric an event is defined as a timestamp, a data value, and an optional quality value. A quality value is sent with the timestamp and data value if provided by the data source.

• The 3,000 stream count and 3,000 events/sec figures are the current commercial limits and not a failure threshold.

• Performance metrics apply to CONNECT and PI Servers as the data destinations.

PI to CONNECT agent

The PI to CONNECT agent enables the transfer of data from PI Servers to CONNECT data services. It supports real-time and historical data transfer from the PI Data Archive and asset data transfer from the PI Asset Framework. The agent is supported on Windows machines.

Performance metrics

Typical use cases:

• In typical usage, most streams are not frequently accessed. In these cases, having 500,000 events/sec uncached streams does not significantly affect performance.

Additional notes:

• CONNECT does not enforce a hard stream limit. Performance may degrade when accessing streams of more than 80,000 events/sec frequently.

CONNECT to PI agent

The CONNECT to PI agent enables the transfer of real-time and historical data from CONNECT data services to the PI Data Archive. The agent is supported on Windows machines.

Performance metrics

AVEVA Historian with native replication and CONNECT Data Services APIs

This configuration uses asynchronous replication to CONNECT Data Services from AVEVA Historian, supporting large-scale tag replication with stable performance.

Test bed description

• System Platform and AVEVA Historian configured with 2 million tags.

• 7-day continuous test period.

Performance metrics

Additional notes

• CPU and memory usage remained stable throughout testing.

Historian with Adapter framework

Test bed description

• Stand-alone agent using Adapter Framework and OMFv1.2.

• 7-day test period with backfill enabled.

Performance metrics

Additional notes

• Limitations:

  • Large data sends may overwhelm the agent causing buffer overwrite and data loss.

  • Historical data sets may include redundant data. Filtering is recommended.

  • Updates with unchanged timestamps and qualities, but non-increasing values, are ignored by CONNECT Data Services.

• Configuration guidance:

  • Increase the buffer size or implement flow control between agent and source system.

Enterprise SCADA

Enterprise SCADA integration uses the Adapter Framework and OMFv1.2 to replicate data to PI and CONNECT.

Test bed description

• Stand-alone agent using Adapter Framework and OMFv1.2.

• No backfill enabled.

Additional notes

• Flow control issues have been reported.