Understand PI Analysis Service settings

There are several default settings that can be modified to improve PI Analysis Service performance. These settings can be enabled or disabled depending on your organization's particular needs, and are listed below. For information on how to configure the parameters covered in this topic, see View or modify analysis service configuration.

Note: Changes to the AutoBackfillingEnabled, MaximumAllowedAutoBackfillingSpanInHours or NumberParallelDataPipes settings require a restart of PI Analysis Service to take effect.

List of configuration settings for PI Analysis Service

The following configuration settings can be modified for the PI Analysis Service from the Configuration window.

• AutoBackfillingEnabled

  This property enables or disables automatic backfilling of gaps in data that result from periods when PI Analysis Service is not active. Automatic backfilling is enabled by default. In some cases, it may not be needed (in a test environment, for example). Enter False to disable this property. If you update this setting, you must restart PI Analysis Service.

• AutoRecalculationEnabled

  This property enables or disables automatic recalculation of analyses. Automatic recalculation executes analyses when a new input with a time stamp before the latest evaluated trigger is received. This useful feature automatically recalculates late-arriving inputs. Automatic recalculation is enabled by default. Enter False to disable this property.

  Note: Although automatic recalculation is enabled globally by setting this parameter, you need to opt in for an automatic recalculation at individual analysis (template) level. For more information, see Configure automatic recalculation of analyses and analyses templates.

• AutoRecalculationIgnoreTimeInSeconds

  This setting allows users to specify a duration whereby any input events with time stamps within this duration and older than the latest trigger time are ignored for automatic recalculation. The default value for this property is 30 seconds. Setting this parameter can be useful for filtering out noises in some analysis inputs that are not desirable for automatic recalculation, for example, if they are known to be delayed.

• AutoRecalculationMinWaitTimeInSeconds

  This setting controls how long analysis service will wait before queuing an analysis for automatic recalculation upon receiving an out-of-order triggering event. A valid range for this setting is 30 to 900 seconds, with a default of 60 seconds. The wait time is reset with the arrival of new out-of-order triggering events. Regardless of this setting, any analyses that have had an out-of-order triggering event will be queued for recalculation after 900 seconds.

• CacheTimeSpanInMinutes

  PI Analysis Service dynamically determines data cache time span for AF attributes used as inputs in calculations. For example, for analyses calculating hourly summary, PI Analysis Service would automatically cache an hour worth of data for required input attributes. This ensures that data required for calculations is available in the data cache, while unnecessary data can be trimmed. Manually set CacheTimeSpanInMinutes only if the time span cannot be determined by calculation configuration. A longer time setting means that any input data needed for calculations is available in the data cache, but it uses more system resources. For calculations that require earlier data (for example, from an hour earlier), you can specify a longer period for this setting to keep data cached, thus avoiding calls to Data Archive. The maximum time span that would be cached is 7 days and any data in the time span beyond 7 days would be trimmed. Note that the MaxCacheEventsPerAttribute setting takes precedence over this setting, so both settings may have to be adjusted simultaneously.

• CalculationWaitTimeInSeconds

  A period of time to wait before evaluating analyses that are running in real-time. This wait time can be used to compensate for late-arriving inputs. For example, if a calculation depends on an interface with a known latency of up to 30 seconds, you might adjust this setting to at least 30 seconds to make sure you get the most current event from the interface. Longer settings reduce the chance of missing calculation inputs caused by latency but increase the delay in getting calculation results. Note that calculations use input values at the trigger time stamp when they are evaluated. The default value for this property is 5 seconds.

• CreateFuturePIPointsForAmbiguousOutputTimes

  A flag that determines the type of PI point to create for newly mapped output attributes that save output history when the time stamp of the output is uncertain. PI Analysis Service attempts to create the PI point type that matches the time stamp of the output, future PI points for output with future time stamps or historical PI points for output with present or past time stamps. If set to False, PI Analysis Service creates historical PI points in cases when the time stamp of the output is ambiguous. If set to True, PI Analysis Service creates future PI points in cases when the time stamp of the output is ambiguous.

• Default to Auto Create PI Points for Template

  When this check box is selected, PI Analysis Service automatically creates PI points for output attributes in analyses derived from analysis templates.

• EvaluationPartitionSize

  Calculations are grouped for efficient evaluation. Analyses based on an analysis template with periodic scheduling, for example, are evaluated together as one calculation group. To speed up evaluation time, very large calculation groups are partitioned into subgroups based on the EvaluationPartitionSize setting; the subgroups can then be evaluated in parallel. The default value for this property is 10,000. For a group of analyses with long individual evaluation times, you may want to specify a smaller partition size. A group of 100 analyses, each performing summary calculations for a day’s worth of data, may evaluate faster as 4 subgroups with a partition size of 25, depending on the CPU resources available on the machine.

• EvaluationsToQueueBeforeSkipping

  If it takes too long to execute a group of calculations, the queue of waiting calculations may become unmanageable. This setting determines when to skip calculations to avoid running out of resources. At the default setting, if more than 50 evaluations are found, PI Analysis Service starts skipping the oldest evaluations to try to return to its normal processing.

• IsLoadSheddingEnabled

  For some applications, accuracy and completeness are critical, and no calculations can be skipped regardless of the delay in completing them. To disable skipping calculations, enter False for this setting.

• IsTelemetryAllowed

  Enter False to stop sending usage data from PI Analysis Service for the Customer Experience Improvement Program. (This does not affect the corresponding setting for program participation you can configure in PI System Explorer.)

• MaxAllowedAutoRecalculationSpanInDays

  The maximum time span in days for which an analysis will be automatically recalculated on receiving an out of order event. Any recalculation time span exceeding this limit will not be considered for automatic recalculation. The default value for this parameter is 180 days.

• MaxCacheEventsPerAttribute

  Once this limit is reached, the oldest events are trimmed from the data cache. Note that this setting takes precedence over the CacheTimeSpanInMinutes setting.

• MaxConcurrentRecalculationRequests

  The maximum number of backfilling or recalculation operations that PI Analysis Service can perform concurrently. PI Analysis Service dynamically scales the number of threads used for backfilling or recalculation between 1 and the number specified, depending on available resources (CPU and memory) of the machine that is running the service. You can set this parameter to any integer value between 1 and 256. The default value for this parameter is (1/2)*P, where P represents the number of logical processors on the machine that PI Analysis Service is running.

• MaximumAllowedAutoBackfillingSpanInHours

  Specify the maximum downtime (in hours) for the PI Analysis Service to initiate automatic backfilling. Setting the parameter to "0" or a negative number will restore the previously recorded setting for this option. If you update this setting, you must restart PI Analysis Service.

  Note: To disable automatic backfilling, you must disable the AutoBackfillingEnabled parameter.

• MinCacheEventsPerAttribute

  The default setting of 1 ensures that at least one event per attribute remains in the data cache.

• NumberDataWriterThreads

  The number of threads PI Analysis Service uses to write analysis outputs. More threads are useful for high frequency calculations that write values at high rates and for writing to Data Archive on a computer with high network latency (other threads can write values while waiting for a response from Data Archive). Consider increasing the number of threads when analyses skip many evaluations or analyses have high evaluation lag. More threads can help if data writes cause a bottleneck. You might set the number of threads to the number of processors on the computer where PI Analysis Service runs.

• NumberEvaluationThreads

  The number of threads PI Analysis Service uses for evaluating analyses concurrently. You can set this parameter to any integer value between 1 and 256. The default value for this parameter is P, where P represents the number of logical processors on the machine that PI Analysis Service is running. Consider increasing the number of evaluation threads when analyses are evaluating with high lag or skipping evaluations, but PI Analysis Service is not fully utilizing available CPU resources. This may happen when you have a lot of analyses with summary calculations that are triggered frequently. Having a larger number of evaluation threads is helpful for analyses that require frequent access to massive amount of input attribute data from outside the data cache. Note that increasing the number of threads for reducing lag or skipped evaluations may not help in cases where PI Analysis Service is already using high CPU.

• NumberParallelDataPipes

  Attributes receive updates from data sources, including Data Archive, through data pipes. Using a single data pipe to update a large number of attributes may take an unacceptably long time. You can increase the number of data pipes which can operate simultaneously to decrease update time. This can be especially useful for calculations with high-frequency data. Keep in mind, however, that increasing the number of data pipes increases the load on Data Archive. If you update this setting, you must restart PI Analysis Service.

• RuntimeStorageFolderPath

  Specify the location of the directory that contains PI Analysis Service run-time information such as user backfilling and calculation groups . By default, PI Analysis Service stores run-time information in \\ProgramData\OSIsoft\PIAnalysisNotifications.

• Suggested PI Point Name

  You can modify the naming pattern here with a substitution parameter. For a complete list, see List of PI AF substitution parameters. Note, however, that some parameters may not be meaningful in the context of a PI point name.

  You can specify PI Point attributes as well as PI Point names. For more information on the attribute syntax, see the examples in Configuration strings for PI point data references. Do not enter the pointtype and future attributes as they are automatically set by asset analytics.