Compression Tuning

Overview

Compression Tuning helps administrators balance two competing goals:

• Preserve the process behavior, peaks, valleys, and movement that users need.
• Avoid storing archive events that do not add meaningful information.

A tuning run observes selected PI Points, compares a higher-detail reference signal with the events retained in the archive, and recommends exception and compression settings. Recommendations are not applied automatically. An administrator reviews the results and decides whether to apply, skip, or exclude each PI Point.

Why compression matters

PI Data Archive compression controls which incoming values become stored archive events. If compression is too loose, noise and redundant updates increase archive volume, network traffic, storage use, and the amount of data that scans and visualizations must process. If compression is too aggressive, meaningful peaks, steps, and process movement can disappear from the archive.

The goal is not to retain the fewest events. It is to filter redundant noise while preserving the behavior that operators and analyses need. Compression Tuning uses observed reference data, signal safeguards, and a target Data Fidelity to estimate that balance. It leaves the final decision with an administrator because the correct trade-off depends on the process and the importance of the signal.

Compression Tuning requires administrator permissions. The PI Data Archive target must also allow compression tuning writes before PI Nexus+ can temporarily adjust observation settings or apply final recommendations.

The page is organized into four tabs:

• Runs: create and monitor tuning runs, then open a run's detailed review.
• Recommendations: review recommendations across runs and perform bulk actions.
• Auto Tune: configure recurring, bounded batches for large PI Point populations.
• Point Exclusions: manage PI Points that should not be included in future tuning runs.

How a tuning run works

A normal tuning workflow is:

1. Define the scope and observation settings, then confirm the temporary-settings warning.
2. Queue the run. Processing continues in the background; the browser does not need to remain open.
3. PI Nexus+ saves each eligible point's original exception and compression settings.
4. During observation, it temporarily lowers only exception deviation so more source updates reach the PI Snapshot. It listens for Snapshot updates as they arrive; it does not repeatedly poll every PI Point.
5. The existing archive compression settings remain unchanged during observation, allowing the reference signal to be compared with current archive behavior.
6. When observation is complete, PI Nexus+ restores the original exception settings before analysis.
7. The analyzer compares the reference and archive streams, tests candidate settings, classifies the result, and produces a recommendation where a safe candidate exists.
8. The run becomes ready for review. An administrator compares the trends and explicitly applies, skips, or excludes each point.

PI Nexus+ cannot create source events that do not exist. A point that receives one update per hour can contribute only about one reference sample per hour. Slow, inactive, or interrupted processes may therefore need a longer observation and may otherwise return Insufficient data.

Create a tuning run

Open Compression Tuning > Runs and select Create tuning run.

• Run name is optional, but a meaningful name helps identify a maintenance window, plant area, or tuning objective later.
• Scope can be PI Point Rule or Selected PI Points. A rule scope follows the rule's eligible inventory; the dialog shows the candidate count. A selected-point scope uses the in-dialog picker, so points can be searched and added without leaving the run setup.
• Adaptive samples (Recommended) lets each point finish independently after it reaches its sample target and minimum duration. This is usually the best choice when points have different update rates.
• Fixed window observes every selected point for the same duration. Use it when all points must be compared over a controlled common interval.

For Adaptive samples, choose Quick (1,000 samples), Recommended (2,000 samples), or High confidence (5,000 samples). The defaults are 1 hour minimum and 7 days maximum. If the maximum is reached first, valid samples are analyzed when at least 20 are available; points below that minimum remain Insufficient data. Fixed observation is used when a common controlled interval is more important than per-point completion.

Select a Tuning profile to control concurrency and historian load: Conservative uses the lowest concurrent load, Balanced is the normal default, Fast uses more concurrency, Legacy preserves legacy behavior, and Custom is available when custom concurrency is exposed. The safety summary estimates points, observation waves, active tags, and duration. Choose a lower-load profile or split a large scope when the archive is busy. Before queuing, confirm that temporary exception/compression settings may be adjusted and restored; Queue is enabled only after the scope and safety confirmation are valid.

Auto Tune Plans

Use Compression Tuning > Auto Tune when a large PI Point population should be processed as smaller recurring batches. A plan can follow a PI Point Rule as its membership changes, or keep an immutable snapshot created from PI Points selected on the PI Points page or a PI Point detail view. Existing plans can be edited without clearing their per-point coverage.

Plans are first-pass-only by default: safely completed PI Points are covered once for that plan. Enable retuning and choose a cooldown when covered points should become eligible again. Plan cards show covered versus scoped points, retry state, and the next scheduled action. Infrastructure failures retry the exact same batch up to three times; individual point failures remain eligible on a later pass, while restore-safety failures pause the plan. Auto Tune only creates recommendations—final settings always require manual review and application.

Runs execute in the background. Large scopes are processed in bounded waves, so the browser does not need to remain open.

Run list and Run Detail

The Runs tab shows run status, review counts, runtime, and actions. Use the status cards to filter All, In Progress, Ready, Completed, or Failed runs. A run can be renamed, canceled while active, archived after it reaches a terminal state, or reopened for review. Enable Show archived runs when historical runs are needed.

Opening a run displays its lifecycle tabs and review counts:

• Ready: analysis completed and the result is available for review.
• Pending: the PI Point is queued, being prepared or observed, waiting for analysis, or being analyzed. During Adaptive observation, the row can show valid-sample progress.
• Applied / Resolved: the recommendation was applied, or the point was otherwise resolved for this run. The label can appear as Resolved when the point was superseded or excluded after processing.
• Skipped: the PI Point was not processed, for example because it was excluded or unsupported.
• Failed / canceled: processing did not complete successfully or the run was canceled.

Ready results can be filtered as All ready, Apply, Investigate, or No action. The table shows the PI Point, analysis classification, recommendation, estimated fidelity, and actions. The page distinguishes actions on selected rows or the loaded list from actions on every point matching the current filter; check that scope before using a bulk action. Large result sets are loaded in pages.

If a point needs closer inspection, use its action menu to compare the trend, open the PI Point detail, apply compression, skip the point for this run, or exclude it from future runs.

How Reference Data Is Collected

Before observation begins, PI Nexus+ reads and stores the current exception and compression settings for every eligible PI Point.

During observation, PI Nexus+ temporarily lowers only the exception deviation to a conservative near-zero value. This allows more source updates to reach the PI Snapshot and creates a higher-detail reference signal. PI Nexus+ listens for Snapshot updates as they arrive; it does not repeatedly poll every PI Point.

This reference is sometimes described as raw data. More precisely, it is the highest-detail stream PI Nexus+ can observe from the updates delivered by the source system during the run.

The existing archive compression settings remain unchanged during observation. The PI Data Archive continues storing events using the current compression configuration, which allows PI Nexus+ to compare the reference signal with the archive behavior being evaluated.

When observation is complete, PI Nexus+ restores the original exception settings before analyzing the results. The comparison uses:

• Snapshot: higher-detail reference updates observed during the run.
• Archive: events retained by the current settings over the same period.
• Recommended: a simulation of the events that the suggested settings would retain.
• Custom: an optional simulation using values entered by the administrator.

PI Nexus+ cannot create source events that do not exist. A PI Point that receives one update per hour can contribute only about one reference sample per hour. Slow, inactive, or interrupted processes may therefore need a longer observation and may otherwise return Insufficient data.

Original exception settings are also restored when a run is canceled or encounters a handled failure. If restoration fails, the failure remains visible so an administrator can retry the restore or correct the PI Point manually.

Observation Modes

Adaptive samples (Recommended) allows each PI Point to finish independently after it reaches both a sample target and the minimum observation duration. If the maximum duration arrives first, PI Nexus+ analyzes the valid samples collected so far when at least 20 are available and reports lower confidence where appropriate. Only points with fewer than 20 valid samples remain insufficient. This is the preferred mode when a scope contains signals with different update rates.

Available sample targets are:

The default minimum observation duration is 1 hour and the default maximum is 7 days. If a PI Point does not reach its sample target before the maximum duration, its settings are restored and the collected data is analyzed when at least 20 valid samples are available. Points below that minimum are reported as Insufficient data.

Fixed window observes all selected PI Points for the same duration. It is useful for controlled tests, known high-rate signals, or cases where a fixed start and end time are required.

Data Fidelity

Data Fidelity estimates how closely a compressed series reproduces the shape and movement of the observed reference signal. It is not a measurement of sensor accuracy, sensor calibration, or PI Point data quality.

A target of 97% is recommended for most process PI Points. A lower target can reduce the number of retained archive events, but it increases the risk of losing smaller process movements. A higher target preserves more detail, but normally stores more events.

The comparison also shows archive count, archive ratio, estimated bandwidth, and estimated storage. Use these values to understand the storage impact, but review Data Fidelity and the visual trend before applying a recommendation. A large reduction in events is not beneficial when important signal behavior is lost.

Result Classifications

Undercompressed and Overcompressed risk describe different problems. Undercompressed means the archive is probably retaining excess events. Overcompressed risk means useful process detail may be discarded. A high overall fidelity value does not always rule out overcompression risk because smaller but important movements may still be hidden.

Signal metadata such as measurement type and behavior (for example slow/smooth, peak-sensitive, step/event, noisy/high-frequency, or general analog) helps the analyzer apply safeguards. It is supporting context, not a replacement for process knowledge.

Compare the trend

Open a PI Point result's Compare trend action. The comparison dialog shows the same time range as four possible series:

• Snapshot (reference): the higher-detail updates observed while exception deviation was temporarily lowered. It is not a magical raw historian stream; it contains only updates delivered by the source during the observation period.
• Archive: events retained under the point's configured exception and compression settings during that period.
• Recommended: a simulation of the candidate settings applied to the reference stream. It estimates fidelity and event reduction; it is not written until an administrator applies it.
• Custom: a live simulation using values entered in the dialog.

Use the legend to show or hide series. Use Pan to move through the time range, Box zoom to inspect a peak, step, or noisy interval, and Reset to return to the full view. Compare visual behavior before optimizing event count: the best candidate removes redundant noise without hiding meaningful process movement.

The dialog compares configured and candidate values such as Exception Deviation, Compression Deviation, Data Fidelity, archive event count and ratio, and estimated bandwidth/storage. A result can show No change recommended when the current settings are already suitable. A candidate with a guardrail warning requires additional review before it can be applied.

Apply, customize, skip, or exclude

The available review actions are:

• Apply recommended: write the tested recommended exception and compression settings after reviewing the trend and safeguards. If the candidate has a warning, the dialog presents the guarded apply action so the administrator can make an explicit decision.
• Apply custom: enable Custom, then either choose a target fidelity or enter custom Exception Deviation and Compression Deviation values. Review the recalculated trend and metrics, then apply the custom candidate. Editing the custom values only changes the simulation; Apply custom is what writes the final settings.
• Skip: mark the PI Point as handled for this run without changing its PI settings. Use this when the data is not representative or when you are unsure; the point remains eligible for a later tuning run.
• Exclude: prevent the PI Point from being included in future tuning runs until an administrator allows it again in Point Exclusions. Use this for points that should be permanently managed outside automated tuning. Excluding a point does not change its existing compression settings.

The previous and new settings, acting user, timestamp, and result are recorded in the tuning audit and normal Change History. If an apply fails because of permissions, connectivity, or a missing PI Point, the error remains visible for administrator action.

Use the Recommendations tab to review available recommendations across tuning runs. Use Point Exclusions to manage PI Points that should not be included in future runs.