This reference lists the classifications, observation modes, limits, profiles and statuses of Compression Tuning. The User Guide chapter Compression Tuning links here.
Overview
This reference lists the classifications, observation modes, limits, profiles and statuses of Compression Tuning. The User Guide chapter Compression Tuning links here.
Result classifications
Classification (as shown)
Meaning
Review group
Suggested response
Undercompressed
The point stores more archive events than needed while already preserving the signal
Apply when the candidate passes every guardrail, otherwise Investigate
Compare the trends; apply the larger deviations
Overcompressed Risk
The current settings may hide real process movement. A high overall fidelity does not rule this out
As above
Compare Snapshot and Archive; apply the more conservative recommendation
Good
The current settings balance signal and volume; the tested alternative brings no material safe reduction
No action
Skip when reviewed
Hard To Compress
The signal is so noisy or busy that reducing events would remove detail. Shown as Limited compression savings when the safeguards pass only with a dense archive
Investigate
Check sensor noise, vibration or instrumentation first
Limited data
Samples were analyzed, but the evidence is too thin to trust the calculated settings
Investigate
Observe longer, during representative operation
Flatline
The signal shows little or no movement
Investigate
Check whether the process is stable or the point is inactive or frozen
Zero/span issue
Observed values do not fit the point's zero and span
Investigate
Correct zero and span before tuning
Insufficient Data
Fewer than 20 valid samples
Investigate
Run again with a longer observation
Unsupported
Non-numeric type or unsupported metadata
Not listed
Do not tune with Compression Tuning
Excluded
Excluded from future runs
Not listed
Allow tuning again on Point Exclusions
Only Undercompressed and Overcompressed Risk can be applied in bulk. Hard To Compress, Limited data, Flatline, Zero/span issue and Insufficient Data, and actionable results whose candidate failed a guardrail, can only be applied one point at a time from the comparison dialog with Apply tested candidate.
When several runs tuned the same point, the newest run with a valid result supplies the current recommendation; older unresolved results become Superseded.
Observation modes and limits
Setting
Adaptive samples (Recommended)
Fixed window
Point finishes
When it has reached the sample target and the minimum duration
After the window
Sample target
Quick 1,000, Recommended 2,000 (default), High confidence 5,000
Not used
Duration
Minimum 1 hour (default); maximum default 168 hours (7 days), up to 720 hours
1 to 24 hours, default 1
At the maximum
The valid samples are analyzed when there are at least 20; otherwise Insufficient Data
Applied, or resolved (superseded or excluded after processing)
Skipped
Skipped for this run, excluded or unsupported
Failed / canceled
Failed, Apply failed, Restore failed, Canceled
Safety
Before observation, the original exception and compression settings of every point are saved.
Only exception deviation is lowered during observation; compression settings are not touched.
Settings are restored before analysis, and also when a run is canceled or fails. A failed restore stays visible and is never cleaned up automatically.
Every applied change records the previous and new settings, user and time in the tuning audit and in Change History.
Auto Tune
Item
Behaviour
Membership
Follows a PI Point Rule, or a fixed selection made on the PI Points page or a point's dialog
Coverage
Each point once per plan; enable retuning and a cooldown (default 90 days) to repeat
Batch size
At most 10,000 points
Failures
An infrastructure failure retries the same batch up to 3 times (after 15 minutes, 1 hour and 4 hours); a point failure is retried on a later pass; a restore failure pauses the plan
Capacity
Open actionable recommendations plus points reserved by running batches. Warning from half the maximum; Held (no new batches) at the maximum, 100,000 by default (10,000 to 1,000,000); resumes below 80 % of it
